From ronnyk at CS.Stanford.EDU  Sat Dec  3 10:26:20 1994
From: ronnyk at CS.Stanford.EDU (Ronny Kohavi)
Date: Sat, 3 Dec 1994 23:26:20 +0800
Subject: MLC++: Machine Learning Utilities Available
Message-ID: <9412040726.AA14178@starry.Stanford.EDU>


			   MLC++ Utilities                
                           _______________


MLC++ is a Machine Learning library of C++ classes being developed at
Stanford.  More information about the library can be obtained at URL
http://robotics.stanford.edu:/users/ronnyk/mlc.html.

We are now releasing the object code for some utilities written using MLC++.
These will run on Suns, either on SunOS or Solaris.  


Included in the current release are the following induction algorithms:

1. Majority (baseline).
2. Basic ID3 for inducing decision trees.  The output can be sent
     to a mail server to get a postscript picture of the resulting
     tree.  Very useful for looking at the final tree and for teaching.
3. Basic nearest neighbor.
4. Decision Table.
5. Interface to C4.5 (for utilities below).
6. Feature subset selection : wraps around any of the above and
     selects a good subset of the features, usually improving
     performance and comprehensibility.
  


Utilities released are:

1. Cross validation : cross validate a file and any of the above induction
   algorithms.  Allows regular or stratified CV.
   You can also generate the cross validation files to compare
   your own induction algorithm.
  
2. Learning curve : generate a learning curve for any of the above
   induction algorithms.

3. Project : project the data onto a subset of attributes.

4. Convert : convert nominal attributes to unary encoding or binary
   encoding.


Quick starter guide:
--------------------

The MLC++ utilities are accessible by anonymous ftp to
  starry.stanford.edu:pub/ronnyk/mlc/
There are currently two kits, one for Solaris (MLCutil-solaris.tar.Z)
   and one for SunOS (MLCutil-sunos.tar.Z).

cd <directory>
zcat <kit-name> | tar xvf -

where <directory> is the directory under which the mlc directory will
   be built (e.g., /usr/local), and <kit-name> is the kit appropriate
   for your machine.

The documentation is in utils.ps.
The environment variable MLCDIR must be set to the directory where
    the utilities are installed.

Databases in the MLC++ format, which is very similar to C4.5 format
  can be found in starry.stanford.edu:pub/ronnyk/mlc/db.
  Most datafiles are converted from the repository at UC Irvine.

Questions or help requests related to the utilities should be
    addressed to mlcpp-help at CS.Stanford.EDU

--

   Ronny Kohavi (ronnyk at CS.Stanford.EDU, http://robotics.stanford.edu/~ronnyk)



From marcus at hydra.ifh.de  Mon Dec  5 04:37:52 1994
From: marcus at hydra.ifh.de (Marcus Speh)
Date: Mon, 5 Dec 94 10:37:52 +0100
Subject: Paper - GAUGE THEORY OF THINGS ALIVE AND UNIVERSAL DYNAMICS
Message-ID: <9412050937.AA21444@hydra.ifh.de>

The following paper is now available via anonymous FTP (see below):

GAUGE THEORY OF THINGS ALIVE AND UNIVERSAL DYNAMICS
BY G. MACK, Theoretical Physics, University of Hamburg, Germany

SOURCE FORMAT: 13 pages, latex, uses fleqn.sty (can be removed without harm)
REPORT-NO: DESY 94-184

				ABSTRACT

  Positing complex adaptive systems made of agents with relations
  between them that can be composed, it follows that they can be
  described by gauge theories similar to  elementary
  particle theory and  general relativity. By definition, a universal
  dynamics is able to determine the time development of any such system
  without need for further specification. The possibilities are limited,
  but one of them - reproduction fork dynamics - describes DNA
  replication and is the basis of biological life on earth. It is a
  universal copy machine and a renormalization group fixed point.
  A universal equation of motion in continuous time is also presented.


TO RETRIEVE, connect as USER "anonymous" with PASSWORD "<userid>@<address>" 

a) in Europe, to 
   ftp.desy.de (131.169.30.42) and retrieve one of the files from
   directory "pub/outgoing"

 203657 Dec  5 10:06 DESY_94-184.ps     [PostScript, uncompressed]
  60601 Dec  5 10:06 DESY_94-184.ps.gz  [PostScript, compressed]
  48393 Dec  1 14:48 DESY_94-184.tex    [LaTeX source]

OR in the US to 

b) ftp.scri.fsu.edu (144.174.128.34) and retrieve the LaTeX source
   file from directory "hep-lat/papers/9411":

  48506 Nov 28 07:27 9411059

On the World-Wide Web, you can also access the URL

  http://www.desy.de/pub/outgoing/DESY_94-184.ps.gz  [60KB]



Inquiries and comments should be sent directly to the author,
Prof. G. Mack <i02mac at dsyibm.desy.de>




From wuertz at ips.id.ethz.ch  Mon Dec  5 08:45:23 1994
From: wuertz at ips.id.ethz.ch (Diethelm Wuertz)
Date: Mon, 05 Dec 1994 14:45:23 +0100
Subject: PASE 95: Parallel Applications in Statistics and Economics
Message-ID: <9412051433.AA11623@sitter>


                            5th Anniversary
                          First Announcement
                            
                       International Workshop on 
           Parallel Applications in Statistics and Economics

                     >> Non-linear Data Analysis <<

                         Trier - Mainz, Germany
                      August 29 - September 2, 1995

____________________________________________________________________________
PURPOSE OF THE WORKSHOP:

  The purpose of this workshop is to bring together researchers interested 
  in innovative information processing systems and their applications in the 
  areas of statistics, finance and economics. The focus will be on in-depth 
  presentations of state-of-the-art methods and applications as well as on 
  communicating current research topics. This workshop is intended for 
  industrial and academic persons seeking new ways of comprehending the 
  behavior of complex dynamic systems. The PASE'95 workshop is concerned 
  with but not restricted to the following topics:

   o  Applications in finance, economics, and natural science
   o  Statistical tests for finding deterministic and chaotic behavior
   o  Statistical tests for measuring stability and stationarity
   o  Sampling and retrieving techniques for high frequency data
   o  Modeling and analysis of non-linear multivariate time series
   o  Statistical use of neural networks, genetic algorithms and fuzzy systems

____________________________________________________________________________
WORKSHOP SITE:

  The workshop will be held on a comfortable ship running from Trier to 
  Mainz on the scenic rivers Mosel and Rhine in Germany. The best connection 
  to Trier is to fly to Frankfurt and then to take the train. It will take 
  about 2 1/2 hours. From Mainz there are direct trains to Frankfurt Airport 
  (~ 30 minutes). Detailed traveling possibilities will be announced later.

____________________________________________________________________________
ACCOMMODATION AND REGISTRATION:
 
  All participants will be accommodated on the ship. The price of the 
  accommodation in a double room for 4 nights is Sfr 870.- for the main 
  deck (920.- for the upper deck) including breakfasts, lunches and dinners. 
  The welcome and workshop dinner are also included. The service begins 
  on Tuesday afternoon, August 29th, in Trier with registration and ends 
  on Saturday morning, September 2nd in Mainz. Since the facilities of 
  the ship are dedicated to the workshop the price is unique and no
  reductions can be provided for a shorter participation. Since the number 
  of participants is limited to about 100 persons, early registration is
  recommended. All applications will be handled on the basis "first come 
  first serve". Registration will only be accepted after a pre-payment of 
  Sfr 300.-. The rest of the accommodation expenses of Sfr 570.- or 620.-, 
  respectively, and the workshop fees must be transferred to our account

             PASE Workshop - Dr. Diethelm Wuertz
             Schweizerischer Bankverein, Zurich, Switzerland
             Number of Account: P0-206066.0

  before May 31st 1995. No currencies other than Swiss Francs will be 
  accepted.  In the case of a later remittance we cannot guarantee your
  application.

____________________________________________________________________________
WORKSHOP FEE AND PROCEEDINGS:

  The workshop fee is Sfr 580.- for profit-making companies and Sfr 230.- 
  for others. It includes the Proceedings, published as a Special Issue of 
  the Journal "Neural Network World - International Journal on Neural and 
  Mass-Parallel computing and Information Systems".

____________________________________________________________________________
WORKSHOP SCHOLARSHIPS:

  For students and a limited number of participants from "post-communist" 
  European countries some support and scholarships will be available: 
  please contact the organizers.

____________________________________________________________________________
ABSTRACTS AND DEADLINES:

  Regular abstracts of one page (approximately 30 lines with 60 characters) 
  must be submitted before December 1st, 1994. (For post-deadline papers
  please contact directly: <pase at ips.id.ethz.ch>.)  All contributions will be 
  refereed. The deadline for the full papers (about 10 pages) is April 1st, 
  1995. We cannot guarantee that papers submitted later will be printed in 
  the proceedings. The proceedings will be available at the conference. It 
  is also possible to submit an extended abstract of 4 pages. From these 
  contributions we will select, besides the regular invited speakers, 
  3 further speakers for an invited talk. The deadlines are the same as 
  in the case of regular abstracts. If you plan a soft- or hardware 
  demonstration please contact the organizers. Please send the abstracts 
  and full papers to:

    Hynek Beran, ICS Prag
    Pod vodarenskou vezi 2                FAX:    +42 2 858 57 89
    182 07 PRAGUE 8, Czech Republic       E-mail: pase at uivt.cas.cz

____________________________________________________________________________
ORGANIZATION:

  The Workshop will be organized by the Interdisciplinary Project Center 
  for Supercomputing (ETH Zurich), Olsen & Associates (Research Institute 
  for Applied Economics, Zurich) and the Institute of Computer Science 
  (Academy of Sciences, Prague)

    W.M. van den Bergh, D.E. Baestaens, Erasmus University Rotterdam
    Thilo von Czarnowski, Helaba Frankfurt
    Michel M. Dacorogna, Olsen & Associates Zurich
    Susanne Fromme, SMH Research Frankfurt
    Heinz Muehlenbein, GMD Sankt Augustin
    Gholamreza Nakkaeizadeh, Daimler Benz Forschung Ulm
    Paul Ormerod, Henley Center for Forecasting London
    Emil Pelikan, ICS Czech Academy of Sciences Prague
    Heinz Rehkugler, University of Bamberg
    Marco Tomassini, CSCS Manno
    Dieter Wenger, Swiss Bank Corporation Basel
    Diethelm Wuertz,  IPS ETH Zurich
    Hans Georg Zimmermann, Siemens AG, Munchen

____________________________________________________________________________
FURTHER INFORMATION:

  Further information will be available from anonymous ftp "maggia.ethz.ch"
  (129.132.17.1) or world wide webb "http://www.ips.id.ethz.ch/PASE/pase.html"

____________________________________________________________________________
REGISTRATION FORM - PASE '95:


   ....................................................................
   Mr./Mrs.        First Name        		Surname

   ....................................................................
   Institution or Company

   ....................................................................


   ....................................................................
   Mailing Address

   ....................................................................
   E-mail 

   ....................................................................
   Phone             			 Fax

   o  Fees University Sfr 230.--              o  Main Deck Sfr 870.-
   o  Fees Profit making Company Sfr 580.--   o  Upper Deck Sfr 920.-

   CONTRIBUTION:
   I would like to present a talk and to submit a paper with the 
   following title:

   ....................................................................

   ....................................................................

   ....................................................................

   Please, return this form as soon as possible to:

     Martin Hanf
     IPS CLU B2, ETH Zentrum                
     CH-8092 Zurich, Switzerland            
     E-Mail: pase at ips.id.ethz.ch

____________________________________________________________________________
PASE '95







--------------------------------------------------------------------------
PD Dr. Diethelm Wuertz
Interdisciplinary Project Center for Supercomputing
IPS CLU B3 - ETH Zentrum
CH-8092 ZURICH, Switzerland

Tel. +41-1-632.5567
Fax. +41-1-632.1104
--------------------------------------------------------------------------



From mbrown at aero.soton.ac.uk  Mon Dec  5 05:08:01 1994
From: mbrown at aero.soton.ac.uk (Martin Brown)
Date: Mon, 5 Dec 94 10:08:01 GMT
Subject: neural -probability-fuzzy seminars
Message-ID: <7370.9412051008@aero.soton.ac.uk>

                        ADT '95
                  Date: April 3-5 1995
        Venue: BRUNEL - The University of West London


UNICOM Seminars Ltd., Brunel Science Park,
Cleveland Road, Uxbridge, Middlesex, UB8 3PH, UK
Telephone: +44 1895 256484  FAX +44 1895 813095

ADT 95
Sponsored by:
Department of Trade and Industry (DTI)
Institute of Mathematics and its Applications
IFSA
International Neural Networks Society
European Neural Networks Society
AI Watch
Journal of Neural Computing and Applications
Rapid Data Ltd

Call for Papers, Registration Information, and Abstract Form

An international team of experts describes and explores the
most recent developments and research in Neural Networks,
Bayesian Belief Networks, Modern Heuristic Search Methods 
and Fuzzy Logic. Presentations cover theoretical developments 
as well as industrial and commercial applications.

Main Topics

Neural Networks and their Applications:
	New Developments in Theory and Algorithms; Simulations
	and Hardware; Pattern Recognition; Medical Diagnosis.

Modern Heuristic Search Methods:
	Genetic Algorithms; Classifier Systems; Simulated 
	Annealing; Tabu Search; Hybrid Techniques; Genetic and
	Evolutionary Programming.

Probabilistic Reasoning and Computational Learning:
	Bayesian Belief Networks; Computational Learning;
	Inductive Inference; Statistical Causality in Belief 
	Networks; Applications in Computational Vision, Medicine,
	Legal Reasoning, Fault Diagnosis and Genetics; Model 
	Selection.

Fuzzy Logic:
	Neural Fuzzy Programming; Algorithms and Applications; 
	Fuzzy Control; Fuzzy Data Processing; Fuzzy Silicon Chips.


PROGRAMME COMMITTEE & SPEAKERS:

Neural Networks and their Applications: 
	J G Taylor, King's College London (Chair); I Aleksander,
	Imperial College; M J Denham, University of Plymouth; 
	B Schuermann, Siemens AG; A G Carr, ERA Technology; 
	R Wiggins, DTI. Other Speakers include: S Hancock, Neural 
	Technologies Ltd; J Stonham, Brunel University, 
	J Austin, Univ. of York; B Clements, Clement Neuronics

Modern Heuristic Search Methods:
	V Rayward-Smith, University of East Anglia (Chair); 
	I Osman, University of Kent; C R Reeves, University of
	Coventry; G D Smith, University of East Anglia.
	Other Speakers include: E Aarts, Philips Research, 
	Netherlands; C Hughes, Logica; S Voss, Darmstadt Univ,
	D Schaaffer, Philips, USA; R Battiti, Trento Univ, Italy

Probabilistic Reasoning & Computational Learning:
	A Gammerman, Royal Holloway University of London (Chair);
	J Pearl, UCLA; D Spiegelhalter, Medical Research Council;
	A Dempster, Harvard University; C S Wallace, Monash
	University (Australia); V Vapnik, AT&T Bell Labs; 
	V Uspensky, University of Moscow.  Other Speakers include:
	P Reverberi, Inst. di Analisi dei Sistemi..Rome; M Talamo, Univ 
	degli Studi, Rome; S McClean, Ulster Univ; A Shen, Ind. Univ. 
	of Moscow; J Kwaan, Marconi Simulation

Fuzzy Logic:
	E H Mamdani, Queen Mary & Westfield College (Chair); J Baldwin,
	Bristol University; C Harris, University of Southampton;
	H Zimmerman, RWTH Aachen.Other Speakers include: L Zadeh, 
	Univ of Calif., Berkeley; R Weber, MIT GmbH; R John, De Montfort
	Univ.; H Bersini, IRIDIA, Belgium.


HOW TO CONTRIBUTE:

PRESENTATIONS:
You may wish to present a paper at any of these sessions. 
Please indicate your choice of session with your submission.
Please ensure we receive the title of your proposed talk and a short 
abstract (30-40 words) by 14 January 1995.  The Organising Committee 
will make their selections and we will advise you by 14 February 1995
whether you presentation can be included.  Written papers for the 
proceedings are also welcome. Please include your telephone and fax
numbers and postal address in all correspondence.

PUBLICATIONS:  Preliminary proceedings of this symposium and also
a refereed volume will be published.

Deadlines for Abstracts:  14 January 1995 - acceptance of paper
notified by 14 February 1995.  Extended abstracts (1000-1500) words,
by 15 March 1995.  Submission of papers for refereed proceedings
(3000-5000) words by 31 August 1995.

REGISTRATION: The conference program and registration material will be
available in late February 1995. To ensure receiving your copy,
complete the attached form and return it to ADT 95.

Registration Fee: GBP175 (Academic), GBP350 (Industrial), GBP500-800
(Exhibition), and GBP1000 Commercial Sponsors, preliminary proceedings
included in the fee.  
Late registration after 28 February 1995 will be charged GBP25 extra.


     -------------------------------------------------------------------
Please return the form below to:

ADT 95, UNICOM Seminars Ltd.,
Brunel Science Park, Cleveland Road, Uxbridge, Middlesex, UB8 3PH.
Telephone: +44 1895 256484.
FAX: +44 1895 813095.
Email: adt95 at unicom.demon.co.uk.

    --------------------------------------------------------------------

Name:
Organization:
Position / Department:

Address:




Telephone:
Facsimile:
E-Mail:

Please send me information about the event,

[ ] I wish to present a paper, title and abstract attached

[ ] I am interested in exhibiting

[ ] Send me more information about ADT95


This event is part of Brunel University's Continuing Education
Programme.


If you fail to return this form, you may not receive further
information.


From dayan at cs.TORONTO.EDU  Mon Dec  5 17:40:37 1994
From: dayan at cs.TORONTO.EDU (Peter Dayan)
Date: Mon, 5 Dec 1994 17:40:37 -0500
Subject: The Helmholtz Machine and the Wake-Sleep Algorithm
Message-ID: <94Dec5.174037edt.317@neuron.ai.toronto.edu>

FTP-host: ftp.cs.toronto.edu
FTP-filename: pub/dayan/wake-sleep.ps.Z
FTP-filename: pub/dayan/helmholtz.ps.Z

Two papers about stochastic and deterministic Helmholtz machines are
available in compressed postscript by anonymous ftp from
ftp.cs.toronto.edu - abstracts are given below.

We regret that hardcopies are not available. 

----------------------------------------------------------------------

pub/dayan/wake-sleep.ps.Z

      The Wake-Sleep Algorithm for Unsupervised Neural Networks
				   
  Geoffrey E Hinton, Peter Dayan, Brendan J Frey and Radford M Neal

Abstract:

We describe an unsupervised learning algorithm for a multilayer
network of stochastic neurons.  Bottom-up "recognition" connections
convert the input into representations in successive hidden layers and
top-down "generative" connections reconstruct the representation in
one layer from the representation in the layer above.  In the "wake"
phase, neurons are driven by recognition connections, and generative
connections are adapted to increase the probability that they would
reconstruct the correct activity vector in the layer below.  In the
"sleep" phase, neurons are driven by generative connections and
recognition connections are adapted to increase the probability that
they would produce the correct activity vector in the layer above.


Submitted for publication

----------------------------------------------------------------------

pub/dayan/helmholtz.ps.Z

			The Helmholtz Machine

  Peter Dayan, Geoffrey E Hinton, Radford M Neal and Richard S Zemel

Abstract:

Discovering the structure inherent in a set of patterns is a
fundamental aim of statistical inference or learning.  One fruitful
approach is to build a parameterised stochastic generative model,
independent draws from which are likely to produce the patterns.  For
all but the simplest generative models, each pattern can be generated
in exponentially many ways. It is thus intractable to adjust the
parameters to maximize the probability of the observed patterns, We
describe a way of finessing this combinatorial explosion by maximising
an easily computed lower bound on the probability of the observations.
Our method can be viewed as a form of hierarchical self-supervised
learning that may relate to the function of bottom-up and top-down
cortical processing pathways.


Neural Computation, in press.

----------------------------------------------------------------------


From lazzaro at CS.Berkeley.EDU  Mon Dec  5 17:44:49 1994
From: lazzaro at CS.Berkeley.EDU (John Lazzaro)
Date: Mon, 5 Dec 1994 14:44:49 -0800
Subject: No subject
Message-ID: <199412052244.OAA16160@boom.CS.Berkeley.EDU>


=========================== TR announcement =================================


REMAP: Recursive Estimation and Maximization of A Posteriori Probabilities
---Applications to Transition-based Connectionist Speech Recognition---

by H. Bourlard, Y. Konig & N. Morgan
   Intl. Computer Science Institute
   1947 Center Street, Suite 600
   Berkeley, CA 94704 
   email: bourlard,konig,morgan at icsi.berkeley.edu

ICSI Technical Report TR-94-064

Abstract

In this report, we describe the theoretical formulation of REMAP, an
approach for the training and estimation of posterior probabilities
using a recursive algorithm that is reminiscent of the EM (Expectation
Maximization) algorithm for the estimation of data likelihoods. Although
very general, the method is developed in the context of a statistical
model for transition-based speech recognition using Artificial Neural
Networks (ANN) to generate probabilities for hidden Markov models (HMMs).
In the new approach, we use local conditional posterior probabilities
of transitions to estimate global posterior probabili- ties of word
sequences given acoustic speech data. Although we still use ANNs to
estimate posterior probabilities, the network is trained with targets
that are themselves estimates of local posterior probabilities. These
targets are iteratively re-estimated by the REMAP equivalent of the
forward and backward recursions of the Baum-Welch algorithm to
guarantee regular increase (up to a local maximum) of the global
posterior probability. Convergence of the whole scheme is proven.

Unlike most previous hybrid HMM/ANN systems that we and others have
developed, the new formulation determines the most probable word
sequence, rather than the utterance corresponding to the most probable
state sequence. Also, in addition to using all possible state
sequences, the proposed training algorithm uses posterior probabilities
at both local and global levels and is discriminant in nature.


The postscript file of the full technical report (66 pages) can be copied
from our (anonymous) ftp site as follows:
ftp ftp.icsi.berkeley.edu
username= anonymous
passw= your email address
cd pub/techreports/1994
binary
get tr-94-064.ps.Z


		



From esann at dice.ucl.ac.be  Tue Dec  6 04:33:59 1994
From: esann at dice.ucl.ac.be (esann@dice.ucl.ac.be)
Date: Tue, 6 Dec 1994 11:33:59 +0200
Subject: Neural Processing Letters Vol.1 No.2
Message-ID: <9412061028.AA06065@ns1.dice.ucl.ac.be>

The following articles may be found in the second issue of the "Neural
Processing Letters" journal (November 1994):

- Good teaching inputs do not correspond to desired responses in ecological
  neural networks
     S. Nolfi, D. Parisi
- Recurrent Sigma-Pi-linked back-propagation network
     T.W.S. Chow, G. Fei
- An evolutionary approach to associative memory in recurrent neural
  networks
     S. Fujita, H. Nishimura
- Equivalence between some dynamical systems for optimization
     K. Urahama
- VISOR: schema-based scene analysis with structured neural networks
     W.K. Leow, R. Miikkulainen
- Nonlinear neural controller with neural Smith predictor
     Y. Tan, A. Van Cauwenberghe
- A neural method for geographical continuous field estimation
     D. Pariente, S. Servigne, R. Laurini

Neural Processing Letters in a rapid publication journal, aimed to publish
new ideas, original developments and work in progress in all aspects of the
field of Artificial Neural Networks.  The delay between submission of
papers and publication is maximum about 3 months.  Please don't hesitate to
ask for the instructions for authors, in order to submit your work to
Neural Processing Letters.

We remind you that all information concerning this journal may be found on
the following servers:
- FTP server: ftp.dice.ucl.ac.be
  directory: /pub/neural-nets/NPL
  login: anonymous
  password: your_e-mail_address
- WWW server: http://www.dice.ucl.ac.be/neural-nets/NPL/NPL.html

For any information (subscriptions, instructions for authors, free sample
copies...), you can also directly contact the publisher:
           D facto publications
           45 rue Masui
           B-1210 Brussels
           Belgium
           Phone: + 32 2 245 43 63
           Fax:   + 32 2 245 46 94






_____________________________
D facto publications -
        conference services
45 rue Masui
1210 Brussels
Belgium
tel: +32 2 245 43 63
fax: +32 2 245 46 94
_____________________________




From mario at physics.uottawa.ca  Tue Dec  6 09:18:06 1994
From: mario at physics.uottawa.ca (Mario Marchand)
Date: Tue, 6 Dec 94 10:18:06 AST
Subject: Neural net paper available by anonymous ftp
Message-ID: <9412061418.AA24196@physics.uottawa.ca>

The following paper, which was presented at the NIPS'94 conference,
is available by anonymous ftp at:

ftp://dirac.physics.uottawa.ca/usr2/ftp/pub/tr/marchand


FileName: nips94.ps

Title: Learning Stochastic Perceptrons Under k-Blocking Distributions

Authors: Marchand M. and Hadjifaradji S.

Abstract:
We present a statistical method
that PAC learns the class of stochastic
perceptrons with arbitrary monotonic activation function
and weights $w_i \in \{-1, 0, +1\}$
when the probability distribution that generates
the input examples is member of a family that we call {\em k-blocking
distributions\/}. Such distributions
represent an important step beyond the case
where each input variable is statistically independent
since the 2k-blocking family contains all the Markov
distributions of order k.
By stochastic perceptron we mean a perceptron which,
upon presentation of input
vector $\x$,  outputs~1 with probability  $f(\sum_i w_i x_i - \theta)$.
Because the same algorithm works for any monotonic
(nondecreasing or nonincreasing) activation function $f$
on Boolean domain,
it handles the well studied cases of sigmo\"{\i}ds
and the ``usual'' radial basis functions.

ALSO: you will find other papers co-authored by Mario Marchand in
      this directory. The text file: Abstracts-mm.txt contains a
      list of abstracts of all the papers.

PLEASE: communicate to me any printing or transmission problems.
        Any comments concerning these papers are very welcome.




   ----------------------------------------------------------------
  |  UUU            UUU            Mario Marchand                  |
  |  UUU            UUU            -----------------------------   |
  |  UUU       OOOOOOOOOOOOOOOO    Department of Physics           |
  |  UUU      OOO   UUU      OOO   University of Ottawa            |
  |   UUUUUUUUUUUUUUUU       OOO   150 Louis Pasteur street        |
  |           OOO            OOO   PO BOX 450 STN A                |
  |            OOOOOOOOOOOOOOOO    Ottawa (Ont) Canada K1N 6N5     |
  |                                                                |
  |   ***** Internet E-Mail: mario at physics.uottawa.ca **********   |
  |   ***** Tel: (613)564-9293 ------------- Fax: 564-6712 *****   |
   ----------------------------------------------------------------





From J.Vroomen at kub.nl  Tue Dec  6 12:58:06 1994
From: J.Vroomen at kub.nl (Jean Vroomen)
Date: 6 Dec 94 12:58:06 MET
Subject: PhD research position in Cogn Psy/Comput Linguistics at Tilburg
Message-ID: <MAILQUEUE-101.941206125806.384@fsw_1.kub.nl>

Predoctoral research associate position in
cognitive psychology/computational linguistics
(Ph.D position, 4 year contract)
Tilburg University

A predoctoral position in cognitive psychology/computational
linguistics is available immediately at Tilburg University. The
focus of research will be on the development of a text-to-speech
conversion system with special emphasis on modelling reading ac-
quisition and developmental dyslexia. The development of the model
and the simulations will be conducted together with behavioral
experiments.

In the project, a strong cross-linguistic approach is taken.
Comparative models will be built for English, French, and Dutch. This
approach furthermore includes the development of a computational
measure of the complexity of these writing systems (`orthographic
depth').

Candidates with training in cognitive psychology, cognitive neurop-
sychology combined with computational (connectionist) modelling are
preferred.

Applicants should send a curriculum vitea and a brief description
of fields of interest to:

Professor Beatrice de Gelder
Department of Psychology
Tilburg University
Warandelaan, 2
PO Box 90153
5000 LE Tilburg, The Netherlands
email: b.degelder at kub.nl


From FRYRL at f1groups.fsd.jhuapl.edu  Wed Dec  7 15:30:00 1994
From: FRYRL at f1groups.fsd.jhuapl.edu (Fry, Robert L.)
Date: Wed, 07 Dec 94 12:30:00 PST
Subject: New Neuroprose Entry
Message-ID: <2EE61B8E@fsdsmtpgw.fsd.jhuapl.edu>


     A preprint of a manuscript entitled "Observer-participant models of 
neural computation" which has been accepted by the IEEE Trans. Neural 
Networks is being made available via FTP from the Neuroprose directory. 
 Details for retrieving this article an associated figures follows the 
abstract below:

               ABSTRACT

     Observer-participant models of neural computation

          A model is proposed in which the neuron serves as an information 
channel. Channel distortion occurs through the channel since the mapping 
from input boolean codes to output codes are many-to-one in that neuron 
outputs consist of just two distinguished states. Within the described 
model, the neuron performs a decision-making function.  Decisions are made 
regarding the validity of a question passively posed by the neuron to its 
environment.  This question becomes defined through learning hence learning 
is viewed as the process of determining an appropriate question based on 
supplied input ensembles.  An application of the Shannon information 
measures of entropy and mutual information taken together in the context of 
the proposed model lead to the Hopfield neuron model with conditionalized 
Hebbian learning rules implemented through a simple modification to Oja's 
learning equation.  Neural decisions are shown to be based on a  sigmoidal 
transfer characteristic or in the limit as computational temperature tends 
to zero, a maximum likelihood decision rule.  The described work is 
contrasted with the information-theoretic approach of Linsker.


The paper is available in two files from archive.cis.ohio-state.edu in the
/pub/neuroprose  subdirectory.  The file names are

     fry.maxmut.ps.Z   (compressed postscript manuscript 433927 bytes)
     fry.maxmut_figs.ps.Z (comp. postscript figures 622321 bytes)

Robert L. Fry
Johns Hopkins University/
Applied Physics Laboratory
Johns Hopkins Road
Laurel, MD 20723
robert_fry at jhuapl.edu


From rohwerrj at helios.aston.ac.uk  Wed Dec  7 12:21:49 1994
From: rohwerrj at helios.aston.ac.uk (rohwerrj)
Date: Wed, 7 Dec 1994 17:21:49 +0000
Subject: 2 tech reports on n-tuple classifiers
Message-ID: <22434.9412071721@sun.aston.ac.uk>

FTP-host: archive.cis.ohio-state.edu
FTP-filename: /pub/neuroprose/morciniec.ntup_bench.ps.Z
FTP-filename: /pub/neuroprose/rohwer.ram_bayes.ps.Z

The following two tech reports have been placed on the Neuropose
ftp site:
     archive.cis.ohio-state.edu (128.146.8.52)
  in pub/neuroprose/

Copies are also available on the Aston University ftp site:
      cs.aston.ac.uk (134.151.52.106)
   in pub/docs/

-----------------------------------------------------------------
morciniec.ntup_bench.ps.Z

     The n-tuple Classifier: Too good to Ignore  (11 Pages)
         Michal Morciniec and Richard Rohwer

The n-tuple classifier is compared to 23 other methods on 11
large datasets from the European Community StatLog project.

-----------------------------------------------------------------
rohwer.ram_bayes.ps.Z

 Two Bayesian treatments of the n-tuple recognition method (8 pages)
                           Richard Rohwer

 Two probabilistic interpretations of the n-tuple recognition
 method are put forward in order to allow this technique to
 be analysed with the same Bayesian methods used in connection
 with other neural network models.
-----------------------------------------------------------------

Sorry, no hardcopies available.

Three cheers for Jordan Pollack!!!

Richard Rohwer
Dept. of Computer Science and Applied Mathematics
Aston University
Aston Triangle
Birmingham  B4 7ET
ENGLAND

Tel: (44 or 0) (21) 359-3621 x4688
FAX: (44 or 0) (21) 333-6215
rohwerrj at uk.ac.aston.cs


From mike at PARK.BU.EDU  Tue Dec  6 20:51:49 1994
From: mike at PARK.BU.EDU (Michael Cohen)
Date: Tue, 6 Dec 1994 20:51:49 -0500
Subject: WCNN'95 Call for Papers & Meeting Info
Message-ID: <199412070151.UAA25340@cns.bu.edu>

			WORLD CONGRESS ON NEURAL NETWORKS
				RENAISSANCE HOTEL
			      WASHINGTON, DC, USA
				  July 17-21, 1995
CALL FOR PAPERS!
DEADLINE FOR PAPER SUBMISSION:  February 10, 1995.

Registration Fees:
Category		Pre-registration	Pre-registration	On-Site
			prior to		prior to		
			February 10, 1995	June 16, 1995

INNS Member		$170.00			$250.00			$350.00
Non-member**		$270.00			$380.00			$480.00
Student***		$ 85.00			$110.00			$135.00

**Registration fee includes 1995 membership and a one (1) year subscription to 
the Journal Neural Networks.
***Student registration must be accompanied by a letter of verification from 
department chairperson.  Any student registration received with no 
verification letter will be processed at the higher member or non-member fee,  
depending on current membership status.  Copies of student identification cards 
are NOT acceptable.  This also applies to on-site registration.

ORGANIZING COMMITTEE
John G. Taylor, General Chair
Walter J. Freeman
Harold Szu
Rolf Eckmiller
Shun-ichi Amari
David Casasent

INNS OFFICERS				GOVERNING BOARD
President:  Walter J. Freeman		Shun-ichi Amari
President-Elect:  John G. Taylor	James A. Anderson
Past President:  Harold Szu		Andrew Barto
Secretary:  Gail Carpenter		David Casasent
Treasurer:  Judith Dayhoff		Leon Cooper
Executive Director:  R. K. Talley	Rolf Eckmiller
					Kunihiko Fukushima
					Stephen Grossberg
					Mitsuo Kawato
					Christof Koch
					Bart Kosko
					Christoph von der Mallsburg
					Alianna Maren
					Paul Werbos
					Bernard Widrow
					Lotfi A. Zadeh

PROGRAM COMMITTEE
Shun-ichi Amari				James A. Anderson
Etienne Barnard				Andrew R. Barron
Andrew Barto				Theodore Berger
Artie Briggs				Gail Carpenter
David Casasent				Leon Cooper
Judith Dayhoff				Rolf Eckmiller
Jeff Elman				Terrence L. Fine
Francoise Fogelman-Soulie		Walter J. Freeman
Kunihiko Fukushima			Patric Gallinari
Apostolos Georgopoulos			Stephen Grossberg
John B. Hampshire II			Michael Hasselmo
Robert Hecht-Nielsen			Akira Iwata
Jari Kangas				Bert Kappen
Christof Koch				Teuvo Kohonen
Kenneth Kreutz-Delgado			Clifford Lau
Sam Levin				Daniel S. Levine
William B. Levy				Christof von der Malsburg
Alianna Maren				Lina Massone
Lance Optican				Paul Refenes
Jeffrey Sutton				Harold Szu
John G. Taylor				Brian Telfer
Andreas Weigand				Paul Werbos
Hal White				Bernard Widrow
Daniel Wolpert				Kenji Yamanishi
Mona E. Zaghloul

CALL FOR PAPERS:
PAPERS MUST BE RECEIVED BY FEBRUARY 10, 1995.  Authors must submit 
registration payment with papers to be eligible for the early 
registration fee.  A $35 per paper processing fee must be enclosed 
in order for the paper to be refereed.  Please make checks payable 
to INNS and include with submitted paper.  For review purposes, 
please submit six (6) copies (1 original, 5 copies) plus 3 1/2" disk 
(see instructions below), four page limit, in English.  $20 per page 
for papers exceeding (4) pages (do not number pages).  Checks for 
over length charges should be made out to INNS and must be included 
with submitted Paper.  Papers must be on 8 1/2" x 11" white paper 
with 1" margins on all sides, one column format, single spaced, in 
Times or similar type style of 10 points or larger, one side of paper 
only.  FAX's not acceptable.  Centered at the top of first page 
should be complete title, author name(s), affiliation(s), and mailing 
address(es), followed by blank space, abstract (up to 15 lines), and
text.  The following information MUST be included in an accompanying 
cover letter in order for the paper to be reviewed:  Full title of 
paper, corresponding author and presenting author name, address, 
telephone and fax numbers.  Technical Session (see session topics) 
1st and 2nd choices, oral or poster presentation preferred*, 
audio-visual requirements (for oral presentations only).  Papers 
submitted which do not meet these requirements or for which 
insufficient funds are submitted will be returned.

For the first time, the proceedings of the 1995 World Congress on 
Neural Networks will be distributed on CD-ROM.  The CD-ROM Proceedings 
are included in your registration fee.  Please note that the 4 volume 
book of proceedings will not be printed.  Format of 3 1/2" disk for 
CD-ROM:  Once paper is proofed, completed and printed for review, 
reformat the paper in Landscape format, page size 8" x 5" for CD.  
You may include a separate file with 1 paragraph biographical 
information with your name, company, address and telephone number.  
Presenters should submit their papers in one of the following 
Macintosh or Microsoft Windows formats:  Microsoft Word, 
WordPerfect, FrameMaker, Quark or Quark Professional, PageMaker,
Persuasion, ASCII, PowerPoint, Adobe.PDF, Postscript (text, not 
EPS).  Images can be submitted in TIF or PCX format.  If submitting 
in Macintosh format, you should submit a disk containing a Font 
Suitcase of fonts used in your presentationn to ensure proper match 
or enclose a cover letter with paper indicating a list of fonts 
used in your presentation.  Information published on the CD will 
consist of text, in-line and offset mathematical equations,
black and white images, line art and graphics.  By submitting a 
previously unpublished paper, author agrees to the transfer of the
copyright to INNS for the conference proceedings.  All submitted 
papers become the property of INNS.  Papers and disk to be sent to:

WCNN'95, 875 Kings Highway, Suite 200, Woodbury, NJ 08096-3172 USA.

*When to choose a poster presentation:  more than 15 minutes are 
needed; author does not wish to run concurrently with an invited
talk; work is a continuation of previous publication; author seeks 
a lively discussion opportunity or to solicit future collaboration.

PLENARY SPEAKERS:
Daniel L. Alkon, U.S. National Institutes of Health
Shun-ichi Amari, University of Tokyo
Gail Carpenter, Boston University
Walter J. Freeman, University of California, Berkeley
Teuvo Kohonen, Helsinki University of Technology
Harold Szu, Naval Surface Warfare Center
John G. Taylor, King's College London

SESSION TOPICS:
1.	Biological Vision:  Rolf Eckmiller
2.	Machine Vision:  Kunihiko Fukushima, Robert Hecht-Nielsen
3.	Speech and Language:  Jeff Elman
4.	Biological Sensory-Motor Control:  Andrew Barto, Lina Massone
5.	Neurocontrol and Robotics:  Paul Werbos
6.	Supervised Learning:  Andrew R. Barron, Terrence L. Fine
7.	Unsupervised Learning:  Teuvo Kohonen, Francoise Fogelman-Soulie
8.	Pattern Recognition:  David Casasent, Brian Telfer
9.	Prediction and System Identification:  John G. Taylor, Paul Werbos
10.	Cognitive Neuroscience:  James Anderson, Jeffrey Sutton
11.	Links to Cognitive Science & Artificial Intelligence:  Alianna 
	Maren
12.	Signal Processing:  Bernard Widrow
13.	Neurodynamics and Chaos:  Harold Szu,  Mona E. Zaghloul
14.	Hardware Implementation:  Clifford Lau
15.	Associative Memory:  Christoph von der Malsburg
16.	Applications:  Leon Cooper
17.	Circuits and Systems Neuroscience:  Stephen Grossberg, Lance 
	Optican
18.	Mathematical Foundations:  Shun-ichi Amari, D.S. Levine
19.	Evolutionary Computing, Genetic Algorithms:  Judith Dayhoff

SHORT COURSES:
a.	Pattern Recognition and Neural Nets:  David Casasent, Carnegie 
	Mellon University
b.	Modelling Consciousness:  John G. Taylor, King's College London
c.	Neocognitron and the Selective Attention Model:  Kunihiko 
	Fukushima, Osaka University
d.	What are the Differences & the Similarities Among Fuzzy, Neural,
	& Chaotic Systems:  Takeshi Yamakawa, Kyushu Institute of
	Technology
e.	Image Processing & Pattern Recognition by Self-Organizing Neural 
	Networks:  Stephen Grossberg, Boston University
f.	Dynamic Neural Networks:  Signal Processing & Coding:  Judith 
	Dayhoff, University of Maryland
g.	Language and Speech Processing:  Jeff Elman, University of 
	California-San Diego
h.	Introduction to Statistical Theory of Neural Networks:  Shun-ichi 
	Amari, University of Tokyo
i.	Cognitive Network Computation:  James Anderson, Brown University
j.	Biology-Inspired Neural Networks:  From Brain Research to 
	Applications in Technology & Medicine:  Rolf Eckmiller, 
	University of Dusseldorf
k.	Neural Control Systems:  Bernard Widrow, Stanford University
l.	Neural Networks to Advance Intelligent Systems:  Alianna Maren, 
	Accurate Automation Corporation
m.	Reinforcement Learning:  Andrew G. Barto, University of 
	Massachusetts
n.	Advanced Supervised-Learning Algorithms and Applications:  
	Francoise Fogelman-Soulie, SLIGOS
o.	Neural Network & Statistical Methods for Function Estimation:  
	Vladimir Cherkassky, University of Minnesota
p.	Adaptive Resonance Theory:  Gail A. Carpenter, Boston University
q.	What Have We Learned from Experiences of Real World Applications 
	in NN/FS/GA?:  Hideyuki Takagi, Matsushita Elctrical Industrial Co., Ltd.
r.	Fuzzy Function Approximation:  Julie A. Dickerson, University of 
	Southern Califorrnia
s.	Fuzzy Logic and Calculi of Fuzzy Rules and Fuzzy Graphs:  Lofti 
	A. Zadeh, University of California-Berkeley
t.	Overview of Neuroengineering and Supervised Learning:  Paul 
	Werbos, National Science Foundation

INDUSTRIAL ENTERPRISE DAY:
Monday, July 17, 1995
Enterprise Session:
Chair:	Robert Hecht-Nielsen, HNC, Inc.
Industrial Session:
Chair:	Takeshi Yamakawa, Kyushu Institute of Technology

FUZZY NEURAL NETWORKS:
Tuesday, July 18, 1995
Wednesday, July 19, 1995
Co-Chairs:	Bart Kosko, University of Southern California
		Ronald R. Yager, Iona College

SPECIAL SESSIONS:
Neural Network Applications in the Electrical Utility Industry
Biomedical Applications & Imaging/Computer Aided Diagnosis in 
Medical Imaging
Statistics and Neural Networks
Dynamical Systems in Financial Engineering
Mind, Brain and Consciousness
Physics and Neural Networks
Biological Neural Networks

To obtain additional information (complete registration brochure, 
registration and hotel forms) contact WCNN'95, 875 Kings Highway,
Suite 200, Woodbury, New Jersey 08096-3172 USA, Tele:  (609)845-
1720; Fax:  (609)853-0411; e-mail:  74577.504 at compuserve.com




From jbower at smaug.bbb.caltech.edu  Wed Dec  7 17:13:15 1994
From: jbower at smaug.bbb.caltech.edu (jbower@smaug.bbb.caltech.edu)
Date: Wed, 7 Dec 94 14:13:15 PST
Subject: Call for papers -- CNS*95
Message-ID: <mailman.714.1149540301.24850.connectionists@cs.cmu.edu>

                          CALL FOR PAPERS                        

          Fourth Annual Computational Neuroscience Meeting 
                             CNS*95 
                       July 11 - 15, 1995
                       Monterey, California
                         ................
               DEADLINE FOR SUMMARIES AND ABSTRACTS: 

                   **>> January 25, 1995 <<**
                        ^^^^^^^^^^^^^^^^
        This is the fourth annual meeting of an interdisciplinary
conference intended to address the broad range of research approaches and
issues involved in the field of computational neuroscience.  The last three
annual meetings, in San Francisco (CNS*92), Washington, DC (CNS*93), and
Monterey, California (CNS*94) brought experimental and theoretical
neurobiologists along with engineers, computer scientists, cognitive
scientists, physicists, and mathematicians together to consider the
functioning of biological nervous  systems.  Peer reviewed papers were
presented on a range of subjects related to understanding how nervous
systems compute.

        As in previous years, the meeting will equally emphasize
experimental, model-based, and more abstract theoretical  approaches to
understanding neurobiological computation.  The meeting in 1995 will again
take place at the Monterey Doubletree Hotel and include plenary,
contributed, and poster  sessions.  There will be no parallel sessions and
the full text of presented papers will be published in a proceedings
volume.  The last day of the meeting will be devoted to a series of
informal workshops focused on current issues in computational neuroscience.
 Student Travel funds and Child Day Care will be available.

                       SUBMISSION INSTRUCTIONS:

      With this announcement we solicit the submission of presented papers.
 All papers will be refereed.  Authors should send original research
contributions in the form of a 1000-word (or less) summary and a separate
single page 50-100 word abstract clearly stating their results.  Summaries
are for program committee use only.  Abstracts will be published in the
conference program.  At the bottom of each abstract page and on the first
summary page,  indicate preference for oral or poster presentation and
specify at least one appropriate category and theme from the following 
list:
 
 Presentation categories: 
    A. Theory and Analysis 
    B. Modeling and Simulation
    C. Experimental 
    D. Tools and Techniques


Themes: 
   A. Development 
   B. Cell Biology 
   C. Excitable Membranes and Synaptic   
        Mechanisms 
   D. Neurotransmitters, Modulators, 
        Receptors 
   E. Sensory Systems
       1. Somatosensory 
       2. Visual
       3. Auditory 
       4. Olfactory 
       5. Other systems
   F. Motor Systems and Sensory Motor
     Integration 
   G. Learning and Memory
   H. Behavior 
   I. Cognitive  
   J. Disease     

 
        Include addresses of all authors on the front of the summary and
the abstract including the E-mail address for EACH author. Indicate on the
front of the summary to which author correspondence should be addressed. 
Program committee decisions will be sent to the correspondence author only.
Submissions will not be considered if they lack category information,
separate abstract sheets,  author addresses, or are late.

Submissions can be made by surface mail ONLY by sending 6 copies of the
abstract and summary to:        

        CNS*95 Submissions
        Division of Biology    216-76
        Caltech    
        Pasadena, CA  91125



ADDITIONAL INFORMATION can be obtained by: 

        o Using our on-line WWW information and registration server, URL of:
            http://www.bbb.caltech.edu/cns95.html

        o ftp-ing to our ftp site.
            yourhost% ftp 131.215.137.69
            Name (131.215.137.69:<yourname>): ftp
            Password:  yourname at yourhost.yourside.yourdomain
            ftp> cd cns95
            ftp> ls

        o Sending Email to:
            cns95-registration-info at smaug.bbb.caltech.edu   


CNS*94 ORGANIZING COMMITTEE:

    Co-meeting chair / logistics -   John Miller, UC Berkeley
    Co-meeting chair / program  -    Jim Bower, Caltech

  Program committee:
      Gwen Jacobs,  University of California,  Berkeley
      Catherine Carr, University of Maryland, College Park
      Dennis Glanzman, NIMH/NIH
      Nancy Kopel, Boston University
      Christiane Linster, ESPCI, Paris, France
      Philip Ulinski,  University of Chicago
      Charles Wilson,  University of Tennessee, Memphis

  Regional Organizers:
                          
       Europe- Erik DeSchutter (Belgium)
       Middle East - Idan Segev (Jerusalem)
       Down Under - Mike Paulin (New Zealand)
       South America - Renato Sabbatini (Brazil)
       Asia - Zhaoping Li (Hong Kong)



           ***************************************
                       James M. Bower
                     Division of Biology
                     Mail code:  216-76
                           Caltech
                     Pasadena, CA 91125
                      (818) 395-6817
                      (818) 449-0679 FAX

      NCSA Mosaic laboratory address:  http://www.bbb.caltech.edu/bowerlab
      NCSA Mosaic address for GENESIS:  http://www.bbb.caltech.edu/GENESIS




From lars at ida.his.se  Wed Dec  7 02:31:17 1994
From: lars at ida.his.se (Lars Niklasson)
Date: Wed, 7 Dec 94 08:31:17 +0100
Subject: Lectureship 
Message-ID: <9412070731.AA00135@mhost.ida.his.se>

The following lectureship is available for a connectionist 
with a background in Lingusitics.

===========================================================

Lectureship in linguistics, ref. no. 276-94-40 
University of Skoevde, Sweden.

Deadline for application: Jan. 20th 1995.

The appointment includes both research and lecturing duties.
Applicants  should  posses  a general competence in linguis-
tics, but also be prepared to work within  the  border-areas
of  computer  science and cognitive science, which are areas
where the University is planning to develop  both  education
and  a research competence with special emphasis on linguis-
tics. It is therefore considered a merit  if  the  applicant
has  skills  within  traditional  artificial intelligence or
connectionism. The lectureship will belong to the Department
of  modern  languages,  but will involve lecturing duties at
the Department of computer science.

The University of Skoevde is one of Sweden's  youngest,  but
it  is undergoing a dynamical development. The main focus of
all the University activities is towards the private sector,
industry  and  international  trade,  especially the use and
development of tools for information technology.  Six  areas
of  competence  have been established; engineering, computer
science, cognitive science, economics, language as  well  as
media and arts.

Currently, the education programmes  of  the  Department  of
Modern  Languages include; English, French, Spanish and Ger-
man. In addition, the Department of computer science  offers
a number of 3-year programmes in computer science and cogni-
tive science, as well as a  research  oriented  master  pro-
gramme.  Research is conducted in connectionism, distributed
real-time databases and active databases.

For more information please  contact:  
The vice-chancellor's office  
Lars-Erik Johansson

Email: Lars-Erik.Johansson at sta.his.se






From pollack at cs.brandeis.edu  Wed Dec  7 17:12:09 1994
From: pollack at cs.brandeis.edu (Jordan Pollack)
Date: Wed, 7 Dec 1994 17:12:09 -0500
Subject: compneuro asst prof job
Message-ID: <199412072212.RAA02297@onyx.cs.brandeis.edu>

I recently moved to Brandeis University, near Boston, where there is a
lovely new building housing a new center for complex systems, which
focuses on the brain, the mind, and computation. The center houses
biologists, biochemists, physicists, psychologists, linguists, and
computer scientists.

There is a faculty search which is relevant to the list:

-------

The Volen National Center for Complex Systems at Brandeis University seeks
candidates for a tenure-track assistant professorship to participate in
establishing a Center for Theoretical Neuroscience funded by a grant from
the Alfred P. Sloan Foundation. We are especially interested in theorists
with strong interests and demonstrated commitment to research relevant to
neuroscience.  Candidates are expected to have Ph.D.s (or the equivalent) in
Mathematics, Physics, Computer Science or another quantitative discipline.
The successful candidate will be appointed in the appropriate academic
department/s.  The position carries a reduced teaching load and summer
salary for three years.  Prospective candidates should send a curriculum
vitae, statement of research interests, and arrange for three letters of
recommendation to be sent directly to: Search Committee,
Theoretical/Computational Neuroscientist, Volen Center, Brandeis University,
Waltham, MA 02254. Applications from minority and women are especially
welcome. Consideration of completed applications will start after January 1,
1995, and will continue until the position is filled, although candidates
are strongly encouraged to submit completed applications as soon as
possible.

----

The likeliest candidates for the faculty position will have a body of
work in neuroscience modeling, rather than in neural models of
cognition or engineering. However, there are also several postdocs
available on the same Sloan grant, for theorists or modelers
who want to cross into experimental neuroscience. Contact Eve
Marder at binah.cc.brandeis.edu, Larry Abbott at psy.ox.ac.uk or John
Lisman at binah.cc.brandeis.edu for further info.

Jordan Pollack                           Associate Professor
Computer Science Department              Center for Complex Systems
Brandeis University                      Phone: (617) 736-2713/fax 2741
Waltham, MA 02254                        email: pollack at cs.brandeis.edu

PS Neuroprose is staying at OSU for a while, just with delays in moving files.









From pollack at cs.brandeis.edu  Thu Dec  8 11:51:23 1994
From: pollack at cs.brandeis.edu (Jordan Pollack)
Date: Thu, 8 Dec 1994 11:51:23 -0500
Subject: FLAME:  WCNN'95 - science in the mud pits
Message-ID: <199412081651.LAA01963@garnet.cs.brandeis.edu>

I kept quiet about the first IEEE NN conference in 1987, where every
paper was accepted, and sorted by how well the author's name was
recognized by the local committee, a feast of surplus dollars were
collected in hyperinflated registration fees.

I just never submitted another paper. But these policies are sickening:

>Authors must submit 
>registration payment with papers to be eligible for the early 
>registration fee...Registration fee includes 1995 membership and a 
>one (1) year subscription to  the Journal Neural Networks.
>A $35 per paper processing fee must be enclosed 
>in order for the paper to be refereed.

Reading fees are charged by agents for aspiring science fiction
writers. Linking registration to submission is nothing more than an
admission that all papers are to be accepted.  And a fee of $480 is
paying a lot more than membership, subscription, and a few days of
coffee breaks!

Also, since the size of the intersection of the two sets
   |ORGANIZING COMMITTEE & PLENARY SPEAKERS| 
is 4 instead of 0, I guess a pentium chip was used.

WHERE are the ethical governors of this scientific society?

Jordan Pollack                           Associate Professor
Computer Science Department              Center for Complex Systems
Brandeis University                      Phone: (617) 736-2713/fax 2741
Waltham, MA 02254                        email: pollack at cs.brandeis.edu


From carlos at cenoli1.ulb.ac.be  Thu Dec  8 16:15:38 1994
From: carlos at cenoli1.ulb.ac.be (carlos@cenoli1.ulb.ac.be)
Date: Thu, 8 Dec 94 16:15:38 MET
Subject: paper available: dynamical computation with chaos
Message-ID: <9412081515.AA01490@cenoli5.ulb.ac.be>


FTP-host: archive.cis.ohio-state.edu
FTP-filename: /pub/neuroprose/babloyantz.comput_chaos.ps.Z

The file babloyantz.comput_chaos.ps.Z is now available for copying
from the Neuroprose repository. This is a 13-page long paper with
5 figures. Size is 204239 bytes compressed and 585704 uncompressed.
Sorry, hardcopies not available.

    Computation with chaos: A paradigm for cortical activity

                  A. Babloyantz and C. Louren\c{c}o

   Service de Chimie-Physique, Universit\'{e} Libre de Bruxelles
        CP 231 - Campus Plaine, Boulevard du Triomphe
               B-1050~Bruxelles, Belgium
            e-mail: carlos at cenoli.ulb.ac.be

Abstract: A device comprising two interconnected networks of
oscillators exhibiting spatiotemporal chaos is considered.
An external cue stabilizes input specific unstable periodic orbits
of the first network thus creating an ``attentive'' state.
Only in this state is the device able to perform pattern discrimination
and motion detection. We discuss the relevance of the procedure to the
information processing of the brain.

This paper appeared in Proc. Natl. Acad. Sci. USA, Vol. 91, p. 9027,
13 September 1994.

Thanks to Jordan Pollack for maintaining the archive.

          Carlos Louren\c{c}o

-------------------------------------------------------------------------

    *** Indonesians persist in the racial cleansing of East Timor. ***
       *** Torture and killing of innocents happen daily. ***


From mdavies at psy.ox.ac.uk  Wed Dec  7 11:24:21 1994
From: mdavies at psy.ox.ac.uk (Martin Davies)
Date: Wed, 7 Dec 1994 16:24:21 +0000
Subject: Euro-SPP '95
Message-ID: <9412071624.AA21454@Mac8>

*************************************************************************

EUROPEAN SOCIETY FOR PHILOSOPHY AND SOCIETY

Fourth Annual Meeting
St. Catherine's College, Oxford
Wednesday 30 August - Friday 1 September, 1995

*************************************************************************

FIRST ANNOUNCEMENT AND CALL FOR PAPERS

The Fourth Annual Meeting of the Euro-SPP will begin at 11.30 am on 
Wednesday 30 August and will end at 5.30 pm on Friday 1 September.  

Themes for Invited Symposia include: emotion, attention, artifical life, 
and brain imaging.  The conference will be held in St. Catherine's College, 
Oxford, where accommodation will be available.

We expect to be able to offer an accommodation and meals package for the 
period from Wednesday morning until Friday afternoon for 108 pounds.  In
addition, bed and breakfast accommodation will be available for the Tuesday 
night before the conference, and for the Friday and Saturday nights after the
conference, at a cost of 28 pounds per night.  A limited number of superior 
rooms with private bath will be available at a higher rate.

*************************************************************************

For further information about local arrangements, email:  espp95 at psy.ox.ac.uk.

*************************************************************************

The Society welcomes submitted papers and posters for this meeting.  

Submitted papers and posters are refereed and selected on the basis of quality 
and relevance to both psychologists and philosophers.

Submitted Papers:  Papers should not exceed a length of 30 minutes (about 
12 double-spaced pages).  The full text should be submitted, along with a 300 
word abstract.

Poster Presentations:  Proposals for poster presentations should consist of a 
500 word abstract.  Unless authors indicate otherwise, submitted papers that 
we are not able to accept will also be considered for poster presentation.

The deadline for submission of both submitted papers and poster presentations 
is 20 January 1995.  Please send three copies to:

	Professor Beatrice de Gelder
	Department of Psychology
	Tilburg University
	5000 LE Tilburg
	The Netherlands
or:
	Professor Christopher Peacocke
	Magdalen College
	Oxford  OX1 4AU
	UK

*************************************************************************

For information about membership of the Euro-SPP, email:  espp at kub.nl.

*************************************************************************








From marks at u.washington.edu  Fri Dec  9 20:37:22 1994
From: marks at u.washington.edu (Robert Marks)
Date: Fri, 9 Dec 94 17:37:22 -0800
Subject: Russian Symposium
Message-ID: <9412100137.AA25592@carson.u.washington.edu>


              The 2-nd International Symposium
           on Neuroinformatics and Neurocomputers
                   Rostov-on-Don, RUSSIA
                   September 20-23, 1995

     Organized by Russian Neural Network Society (RNNS)

                          and

 A.B. Kogan Research Institute for Neurocybernetics (KRINC)

                   in co-operation with

     Institute of Electrical and Electronics Enginieers
             Neural Networks Council (IEEE NNC)


                   First Call for Papers

Research in Neuroinformatics and Neurocomputing continued in
Russia after  the research  was deflated  in the west in the
1970's.   The research sophistication in neural networks, as
a   result,  is   quite   advanced   in   Russia.  The first
international  RNNS/IEEE  Symposium, held  in October  1992,
proved  to  be  a  highly  successful  forum  for  a diverse
international  interchange  of   fresh  and  novel  research
results.    The    second   International    Symposium    on
Neuroinformatics   and   Neurocomputers   is   built on this
remarkable success.

The symposium  focus is  on the  neuroscience,  mathematics,
physics,  engineering  and  design  of  neuroinformatic  and
neurocomputing systems.

Rostov-on-Don, the  location of the Symposium, is about 1000
km south  of Moscow  on the  scenic Don  river.   The Don is
commonly identified  as the  boundary between the continents
of Europe  and Asia.   Rostov  is the home of the A.B. Kogan
Research Institute  for  Neurocybernetics  at  Rostov  State
University -  one of  the premier  neural  network  research
centers in Russia.

Papers for  the Symposium  should be  sent  in  CAMERA-READY
FORM, NOT  EXEEDING 8  PAGES in A4  format  to   the Program
Committee Co-Chair Alexander A. Frolov.   Two copies  of the
paper should  be submitted.

The deadline for submission is 15 MARCH, 1995.

Notification of acceptance will be sent on or before 15 May,
1995.




                    SYMPOSIUM COMMITTEE

GENERAL CHAIR

     Witali L. Dunin-Barkowski, Dr. Sci.,
     The 2-nd International Symposium
     on Neuroinformatics and Neurocomputers,
     Symposium Chair
     President of the Russian Neural Network Society,
     A.B. Kogan Research Institute for Neurocybernetics
     Rostov State University
     194/1 Stachka avenue,
     344104, Rostov-on-Don, Russia
     Tel: +7-8632-28-0588,
     Fax: +7-8632-28-0367
     E-mail: wldb at krinc.rostov-na-donu.su


PROGRAM COMMITTEE CO-CHAIRS

     Professor Alexander A. Frolov,
     The 2-nd International Symposium on Neuroinformatics and
        Neurocomputers,
     Program Co-Chair
     5a Butlerov str.
     Higher Nervous Activity and Neurophysiology Institute
     Russian Academy of Science
     117220, Moscow, RUSSIA.

     Professor Robert J. Marks II
     Program Co-Chair
     The 2-nd International Symposium on Neuroinformatics and
        Neurocomputers
     University of Washington
     Department of Electrical Engineering
     c/o 1131 199th Street S.W., Suite N
     Lynnwood, WA 98036-7138
     WA, USA.

Other information is available from the Symposium Committee.





From john at dcs.rhbnc.ac.uk  Sat Dec 10 07:18:37 1994
From: john at dcs.rhbnc.ac.uk (john@dcs.rhbnc.ac.uk)
Date: Sat, 10 Dec 94 12:18:37 +0000
Subject: Technical Report Series in Neural and Computational Learning
Message-ID: <28465.9412101218@platon.cs.rhbnc.ac.uk>


The European Community ESPRIT Working Group in Neural and Computational 
       Learning Theory (NeuroCOLT): one new report available

----------------------------------------
NeuroCOLT Technical Report NC-TR-94-011:
----------------------------------------
Valid Generalisation from Approximate Interpolation

by  Martin Anthony, Department of Mathematics, The London School of Economics
    Peter Bartlett, Research School of Information Sciences and  Engineering,
       Australian National University
    Yuval Ishai, Department of Computer Science, Technion
    John Shawe-Taylor, Department of Computer Science, Royal Holloway, 
       University of London

Abstract:
Let $\H$ and $\C$ be  sets of functions from domain $X$ to the reals. We say
that $\H$ validly generalises $\C$ from approximate interpolation if and only if
for each $\eta>0$ and $\epsilon, \delta \in (0,1)$ there is a number
$m_0(\eta,\epsilon, \delta)$ such that for any function $t \in \C$ and any
probability distribution $P$ on $X$,  if $m \ge m_0$ then with $P^m$-probability
at least $1-\delta$, a sample $\vx =(x_1, x_2, \dots, x_m) \in X^m$ satisfies
$$\forall h \in \H, \,
|h(x_i) - t(x_i)|< \eta, \,(1 \le i \le m) \Longrightarrow
 P(\{x: |h(x) -t(x)| \ge \eta\}) < \epsilon.$$
We find conditions that are necessary and sufficient for $\H$ to validly
generalise $\C$ from approximate interpolation, and we obtain bounds on the
sample length $m_0(\eta, \epsilon, \delta)$ in terms of various parameters
describing the expressive power of $\H$.

----------------------------------------
NeuroCOLT Technical Report NC-TR-94-022:
----------------------------------------
Sample Sizes for Sigmoidal Neural Networks
by John Shawe-Taylor, Department of Computer Science, Royal Holloway
       University of London

Abstract:
This paper applies the theory of Probably Approximately Correct (PAC)
learning to feedforward neural networks with sigmoidal activation
functions.  Despite the best known upper bound on the VC dimension of
such networks being $O((WN)^2)$, for $W$ parameters and $N$
computational nodes, it is shown that the asymptotic bound on the sample
size required for learning with increasing accuracy $1 - \epsilon$ and
decreasing probability of failure $\delta$ is
$$O((1/\epsilon)(W\log(1/\epsilon) + (WN)^2 + \log(1/\delta)).$$ 
For practical values of $\epsilon$ and $\delta$ the formula obtained for
the sample sizes is a factor $2\log(2e/\epsilon)$ smaller than a naive
use of the VC dimension result would give. Similar results are obtained
for learning where the hypothesis is only guaranteed to correctly
classify a given proportion of the training sample. The results are
formulated in general terms and show that for many learning classes
defined by smooth functions thresholded at the output, the sample size
for a class with VC-dimension $d$ and $\ell$ parameters is
$O((1/\epsilon)(\ell\log(1/\epsilon) + o(\log(1/\epsilon))d +
\log(1/\delta))$.  

-----------------------
The Report NC-TR-94-011 can be accessed and printed as follows 

% ftp cscx.cs.rhbnc.ac.uk  (134.219.200.45)
Name: anonymous
password: your full email address
ftp> cd pub/neurocolt/tech_reports
ftp> binary
ftp> get nc-tr-94-011.ps.Z
ftp> bye
% zcat nc-tr-94-011.ps.Z | lpr -l

Similarly for the other technical reports.

Uncompressed versions of the postscript files have also been
left for anyone not having an uncompress facility.

A full list of the currently available Technical Reports in the Series is held in a file
`abstracts' in the same directory.

Best wishes
John Shawe-Taylor




From cnna at tce.ing.uniroma1.it  Mon Dec 12 09:54:22 1994
From: cnna at tce.ing.uniroma1.it (cnna@tce.ing.uniroma1.it)
Date: Mon, 12 Dec 1994 15:54:22 +0100
Subject: program CNNA-94
Message-ID: <9412121454.AA14007@tce.ing.uniroma1.it>


(Multiple posting - please excuse us if you receive more than a copy)

THIRD IEEE INTERNATIONAL WORKSHOP ON CELLULAR NEURAL NETWORKS AND THEIR 
APPLICATIONS (CNNA-94)
ROME, ITALY, DEC. 18-21, 1994

PRELIMINARY PROGRAM

SUNDAY, DECEMBER 18, 1994 

17.30 Inaugural Session (Hall 1)

17.30 Welcome address
G. Orlandi, Dean, Faculty of Engineering
V. Cimagalli, Chairman, CNNA-94

17.45 Opening address
C. Lau

18.05 Inaugural lecture:
L.O. Chua
The CNN Universal Chip: Dawn of a New Computer Paradigm

19.30 Welcome Cocktail (Room of the Frescoes)

MONDAY, DECEMBER 19, 1994

9.00 Session 1: Theory I (Hall of the Cloister - chairman: L.O. Chua)

9.00 Invited review paper:
T. Roska
Analogic Algorithms Running on the CNN Universal Machine

9.30 G. Yang, T. Yang, L.-B. Yang
On Unconditional Stability of the General Delayed Cellular Neural 
Networks 
9.45 S. Arik, V. Tavsanoglu
A Weaker Condition for the Stability of Nonsymmetric CNNs
10.00 M. P. Joy, V. Tavsanoglu
Circulant Matrices and the Stability Theory of CNNs
10.15 B.E. Shi, S. Wendsche, T. Roska, L.O. Chua
Random Variations in CNN Templates: Theoretical Models and Empirical 
Studies

10.30 Coffee Break (Room of the Frescoes)

11.00 Session 2: Connections with Neurophysiology (Hall of the Cloister - 
chairman: T. Roska)

11.00 Invited lecture:
F. Werblin, A. Jacobs 
Using CNN to Unravel Space-Time Processing in the Vertebrate Retina

11.45 Invited lecture:
J. Hmori
Synaptic Organization of the Thalamic Visual Center (LGN) of Mammals as a 
Basis of CNN Model of Subcortical Visual Processing

12.30 K. Lotz, Z. Vidnynszky, T. Roska, J. Vandewalle, J. Hmori, A. 
Jacobs, F. Werblin
Some Cortical Spiking Neuron Models Using CNN
12.45 T.W. Berger, B.J. Sheu, R. H.-J. Tsai
Analog  VLSI Implementation of a Nonlinear Systems Model of  the 
Hippocampal Brain Region
13.00 A. Jacobs, T. Roska, F. Werblin
Techniques for Constructing Physiologically Motivated Neuromorphic Models 
in CNN

13.15 Lunch (Room of the Frescoes)

14.45 Session 3: Hardware Implementations I (Hall of the Cloister - 
chairman: A. Rodrguez-Vzquez)

14.45 Invited review paper:
A. Rodrguez-Vzquez, R. Domnguez-Castro, S. Espejo
Design of CNN Universal Chips: Trends and Obstacles

15.15 J.M. Cruz, L.O. Chua, T. Roska
A Fast, Complex and Efficient Test Implementation of the CNN Universal 
Machine
15.30 F. Sargeni, V. Bonaiuto
High Performance Digitally Programmable CNN Chip with Discrete Templates
15.45 A. Paasio, A. Dawidziuk, K. Halonen, V. Porra
Digitally Controllable Weights in Current Mode Cellular Neural Networks
16.00 D. Lm, G.S. Moschytz
A Programmable, Modular CNN Cell
16.15 M.-D. Doan, R. Chakrabaty, M. Heidenreich, M. Glesner, S. Cheung
Realisation of a Digital Cellular Neural Network for Image Processing
16.30 R. Domnguez-Castro, S. Espejo, A. Rodrguez-Vzquez, R. Carmona
A CNN Universal Chip in CMOS Technology

16.45 Coffee Break (Room of the Frescoes)

17.15 Session 4: Theory II (Hall of the Cloister - chairman: R.-W. Liu)

17.15 R.-W. Liu, Y.-F. Huang, X.-T. Ling
A Novel Approach to the Convergence of Neural Networks for Signal 
Processing
17.30 E. Pessa, M.P. Penna
Local and Global Connectivity in Neuronic Cellular Automata
17.45 X.-Z. Huang, T. Yang, L.-B. Yang
On Stability of the Time-Variant Delayed Cellular Neural Networks
18.00 J.J. Szczyrek, S. Jankowski
A Class of Asymmetrical Templates in Cellular Neural Networks
18.15 P.P. Civalleri, M. Gilli
A Topological description of the State Space of a Cellular Neural Network
18.30 M. Tanaka, T. Watanabe
Cooperative and Competitive Cellular Neural Networks

TUESDAY, DECEMBER 20, 1994

9.15 Session 5: Learning I (Hall of the Cloister - chairman: J.A. Nossek)

9.15 Invited review paper:
J.A. Nossek
Design and Learning with Cellular Neural Networks

9.45 I. Fajfar, F. Bratkovic
Statistical Design Using Variable Parameter Variances and Application to 
Cellular Neural Networks
10.00 N.N. Aizenberg, I.N. Aizenberg
CNN-like Networks Based on Multi-Valued and Universal Binary Neurons: 
Learning and Application to Image Processing
10.15 W. Utschick, J.A. Nossek
Computational Learning Theory Applied to Discrete-Time Cellular Neural 
Networks
10.30 H. Magnussen, J.A. Nossek
Global Learning Algorithms for Discrete-Time Cellular Neural Networks

10.45 Coffee Break (Room of the Frescoes)

11.15 Session 6: Learning II (Hall of the Cloister - chairman: J. 
Vandewalle)

11.15 H. Magnussen, G. Papoutsis, J.A. Nossek
Continuation-Based Learning Algorithm for Discrete-Time Cellular Neural 
Networks
11.30 C. Gzelis, S. Karamahmut
Recurrent Perceptron Learning Algorithm for Completely Stable Cellular 
Neural Networks
11.45 A.J. Schuler, M. Brabec, D. Schubel, J.A. Nossek
Hardware-Oriented Learning for Cellular Neural Networks
12.00 F. Dellaert, J. Vandewalle
Automatic Design of Cellular Neural Networks by Means of Genetic 
Algorithms: Finding a Feature Detector
12.15 H. Mizutani
A New Learning Method for Multilayered Cellular Neural Networks

12.30 Lunch (Room of the Frescoes)

14.00 Panel Discussion: Trends and Applications of the CNN Paradigm (Hall 
of the Cloister - chairman: L.O. Chua)
Panelists: L.O. Chua, V. Cimagalli, J.A. Nossek, T. Roska, A. Rodrguez-
Vzquez, J. Vandewalle

15.30 Coffee Break (Room of the Frescoes)

16.00 Session 7: Applications I (Hall of the Cloister - chairman: N.N. 
Aizenberg)

16.00 N.N. Aizenberg, I.N. Aizenberg, T.P. Belikova
Extraction and Localization of Important Features on Grey-Scale Images: 
Implementation on the CNN
16.15 K. Slot
Large-Neighborhood Templates Implementation in Discrete-Time CNN 
Universal 
Machine with a Nearest-Neighbor Connection Pattern
16.30 J. Pineda de Gyvez
XCNN: A Software Package for Color Image Processing
16.45 B.E. Shi
Order Statistic Filtering with Cellular Neural Networks
17.00 L.-B. Yang, T. Yang, B.-S. Chen
Moving Point Target Detection Using Cellular Neural Networks
17.15 X.-P. Yang, T. Yang, L.-B. Yang
Extracting Focused Object from Defocused Background Using Cellular Neural 
Networks
17.30 M. Balsi, N. Racina
Automatic Recognition of Train Tail Signs Using CNNs
17.45 A. Kellner, H. Magnussen, J.A. Nossek
Texture Classification, Texture Segmentation and Text Segmentation with 
Discrete-Time Cellular Neural Networks

16.00 Session 8: Applications II (Hall 17 - chairman: S. Jankowski)

16.00 P.L. Venetianer, P. Szolgay, K.R. Crounse, T. Roska, L.O. Chua
Analog Combinatorics and Cellular Automata-Key Algorithms and Layout 
Design
16.15 . Zarndy, T. Roska, Gy. Liszka, J. Hegyesi, L. Kk, Cs. Rekeczky
Design of Analogic CNN Algorithms for Mammogram Analysis
16.30 P. Szolgay, Gy. Erss, A. Katona, . Kiss
An Experimental System for Path Tracking of a Robot Using a 16*16 
Connected Component Detector CNN Chip with Direct Optical Input
16.45 T. Kozek, T. Roska
A Double Time-Scale CNN for Solving 2-D Navier-Stokes Equations
17.00 . Zarndy, F. Werblin, T. Roska, L.O. Chua
Novel Types of Analogic CNN Algorithms for Recognizing Bank-Notes
17.15 B.J. Sheu, Sa H. Bang, W.-C. Fang
Optimal Solutions of Selected Cellular Neural Network Applications by the 
Hardware Annealing Method
17.30 B. Siemiatkowska
Cellular Neural Network for Mobile Robot Navigation
17.45 A. Murgu
Distributed Neural Control for Markov Decision Processes in Hierarchic 
Communication Networks

20.00 Banquet (Restaurant "4 Colonne" - via della Posta Vecchia, 4)

WEDNESDAY, DECEMBER 21, 1994

9.00 Session 9: Spatio-Temporal Dynamics I (Hall of the Cloister - 
chairman: V.D. Shalfeev)

9.00 Invited review paper
V.D. Shalfeev, G.V. Osipov
Spatio-Temporal Phenomena in Cellular Neural Networks

9.30 C.-M. Yang, T. Yang, K.-Y. Zhang
Chaos in Discrete Time Cellular Neural Networks
9.45 R. Dogaru, A.T. Murgan, D. Ioan
Robust Oscillations and Bifurcations in Cellular Neural Networks
10.00 H. Chen, M.-D. Dai, X.-Y. Wu
Bifurcation and Chaos in Discrete-Time Cellular Neural Networks
10.15 M.J. Ogorzalek, A. Dabrowski, W. Dabrowski
Hyperchaos, Clustering and Cooperative Phenomena in CNN Arrays Composed 
of Chaotic Circuits

10.30 Coffee Break (Room of the Frescoes)

11.00 Session 10: Spatio-Temporal Dynamics II (Hall of the Cloister - 
chairman: M. Hasler)

11.00 Invited review paper:
P. Thiran, M. Hasler
Information Processing Using Stable and Unstable Oscillations: A Tutorial

11.30 P. Szolgay, G. Vrs
Transient Response Computation of a Mechanical Vibrating System Using 
Cellular Neural Networks
11.45 P.P. Civalleri, M. Gilli
Propagation Phenomena in Cellular Neural Networks
12.00 S. Jankowski, A. Londei,  C. Mazur, A. Lozowski 
Synchronization Phenomena in 2D Chaotic CNN

12.15 Poster Session (Hall of the Cloister)

S. Jankowski, R. Wanczuk   
CNN models of complex pattern formation in excitable media
Z. Galias, J.A. Nossek
Control of a Real Chaotic Cellular Neural Network
A. Piovaccari, G. Setti
A Versatile CMOS Building Block for Fully Analogically-ProgrammableVLSI 
Cellular Neural Networks
P. Thiran, G. Setti
An Approach to Local Diffusion and Global Propagation in 1-dim. Cellular 
Neural Networks
J. Kowalski, K. Slot, T. Kacprzak
A CMOS Current-Mode VLSI Implementation of Cellular Neural Network for an 
Image Objects Area Estimation
W.J. Jansen, R. van Drunen, L. Spaanenburg, J.A.G. Nijhuis
The AD2 Microcontroller Extension for Artificial Neural Networks
C.-K. Pham, M. Tanaka
A Novel Chaos Generator Employing CMOS Inverter for Cellular Neural 
Networks

13.15 Lunch (Room of the Frescoes)

14.45 Session 11: Hardware Implementations II (Hall of the Cloister - 
chairman: J.L. Huertas)

14.45 R. Beccherelli, G. de Cesare, F. Palma
Towards an Hydrogenated Amorphous Silicon Phototransistor Cellular Neural 
Network
15.00 A. Sani, S. Graffi, G. Masetti, G. Setti
Design of CMOS Cellular Neural Networks Operating at Several Supply 
Voltages
15.15 M. Russell Grimaila, J. Pineda de Gyvez
A Macromodel Fault Generator for Cellular Neural Networks
15.30 P. Kinget, M. Steyaert
Evaluation of CNN Template Robustness Towards VLSI Implementation
15.45 B.J. Sheu, Sa H. Bang, W.-C. Fang
Analog VLSI Design of Cellular Neural Networks with Annealing Ability
16.00 L. Raffo, S.P. Sabatini, G.M. Bisio
A Reconfigurable Architecture Mapping Multilayer CNN Paradigms

14.45: Session 12: Applications III (Hall 17 - chairman: J. Herault)

14.45 G. Adorni, V. DAndrea, G. Destri
A Massively Parallel Approach to Cellular Neural Networks Image 
Processing
15.00 M. Coli, P. Palazzari, R. Rughi
Use of the CNN Dynamic to Associate Two Points with Different 
Quantization Grains in the State Space
15.15 M. Csapodi, L. Nemes, G. Tth, T. Roska, A. Radvnyi
Some Novel Analogic CNN Algorithms for Object Rotation, 3D Interpolation-
Approximation, and a "Door-in-a-Floor" Problem
15.30 H. Harrer, P.L. Venetianer, J.A. Nossek, T. Roska, L.O. Chua 
Some Examples of Preprocessing Analog Images with Discrete-Time Cellular 
Neural Networks
15.45 A.G. Radvnyi
Solution of Stereo Correspondence in Real Scene: an Analogic CNN 
Algorithm
16.00 J.P. Miller,  K.R. Crounse, T. Sziranyi,  L. Nemes,  L.O. Chua, T. 
Roska
Deblurring of Images by Cellular Neural Networks with Applications to 
Microscopy

16.15 Coffee Break (Room of the Frescoes)

16.45 Session 13: Hardware Implementations III (Hall of the Cloister - 
chairman: M. Salerno)

16.45 T. Roska, P. Szolgay, . Zarndy, P.L. Venetianer, A. Radvnyi, T. 
Szirnyi
On a CNN Chip-Prototyping System
17.00 M. Balsi, V. Cimagalli, I. Ciancaglioni, F. Galluzzi
Optoelectronic Cellular Neural Network Based on Amorphous Silicon Thin 
Film Technology
17.15 S. Espejo, R. Domnguez-Castro, A. Rodrguez-Vzquez, R. Carmona
Weight-Control Strategy for Programmable CNN Chips
17.30 S. Espejo, A. Rodrguez-Vzquez, R. Domnguez-Castro, R. Carmona
Convergence and Stability of the FSR CNN Model
17.45 R. Domnguez-Castro, S. Espejo, A. Rodrguez-Vzquez, I. Garca-
Vargas, J.F. Ramos, R. Carmona
SIRENA: A Simulation Environment for CNNs

16.45 Session 14: Applications IV (Hall 17 - chairman: M. Tanaka)

16.45 P. Arena, S. Baglio, L. Fortuna, G. Manganaro
CNN Processing for NMR Spectra
17.00 P. Arena, L. Fortuna, G. Manganaro, S. Spina
CNN Image Processing for the Automatic Classification of Oranges
17.15 S. Schwarz
Detection of Defects on Photolitographic Masks by Cellular Neural 
Networks
17.30 M. Ikegami, M. Tanaka
Moving Image Coding and Decoding by DTCNN with 3-D Templates
17.45 M. Kanaya, M. Tanaka
Robot Multi-Driving Controls by Cellular Neural Networks

GENERAL INFORMATION

Venue: All sessions will be held at the Faculty of Engineering, "La 
Sapienza" University of Rome, vie Eudossiana, 18 - 00184 Rome

Coffee Breaks and lunches are included in the registration fee; they will 
be served in the room of the Frescoes, opposite the main lecture hall. In 
order to receive them, it is necessary to wear the workshop badge.

Banquet will be held at restaurant "4 Colonne", via della Posta Vecchia, 
4, near Piazza Navona, tel. 68307152/68805261. It can be reached on foot 
in about 20 minutes from the Faculty of Engineering, by bus no. 81 or 87 
from via dei Fori Imperiali to Corso Rinascimento, or by taxi. You may 
ask for directions at the registration desk. Please bring the coupon you 
receive at registration.

Communications: You may receive messages by fax, no. +39-6-4742647. Faxes 
should be addressed to your name, with clear statement "participant to 
CNNA-94" in the cover sheet. A UNIX ASCII terminal will be available for 
all participants, with full Internet capabilities. You may receive 
messages at the usual e-mail address of the workshop: 
cnna at tce.ing.uniroma1.it  Subject of messages should include your name.

Public telephones, suitable for local, long-distance, and international 
calls, are available in the Faculty. They accept magnetic cards that can 
be purchased at the registration desk, at tobacconists' and newsstands, 
and in other shops carrying the Telecom symbol.

Bank: Banca di Roma has an agency within the Faculty of Engineering. It 
is open in the morning from 8.25 to 1.35. Several other banks are 
available in the area of the University, also open from 3 to 4pm. Some 
cash dispensers, e.g. the one at Banca di Roma on via Cavour, between 
Hotel Palatino and via dei Fori Imperiali, accept credit cards for cash 
retrieval 24 hours a day. 

Authors are requested to meet with their session chairman 15 minutes 
prior to the session in the same room where it will be held. Time slots 
for regular papers are strictly limited to 15 minutes, including 
discussion. Poster authors should set up posters on 10.30, Wed. Dec. 21, 
and remove them during lunch break.

Demonstrations of working hardware and software for CNNs will be held 
throughout the workshop during breaks. Demonstration authors should bring 
their material for set up on 8am of the day agreed upon, and remove it at 
the end of sessions.


FURTHER INFORMATION
Please contact:
CNNA-94
tel. +39-6-44585836
fax. +39-6-4742647
e-mail cnna at tce.ing.uniroma1.it


From ai at aaai.org  Mon Dec 12 15:36:08 1994
From: ai at aaai.org (AAAI)
Date: Mon, 12 Dec 94 12:36:08 PST
Subject: AAAI 1995 Fall Symposium Series Call for Participation
Message-ID: <9412122036.AA19948@aaai.org>

AAAI 1995 Fall Symposium Series 
Call for Participation

November 10-12, 1995

Massachusetts Institute of Technology 
Cambridge, Massachusetts

Sponsored by the American Association for Artificial Intelligence 
445 Burgess
Drive, Menlo Park, CA 94025 
(415) 328-3123 (voice)
(415) 321-4457 (fax)
fss at aaai.org

The American Association for Artificial Intelligence presents the 1995 Fall
Symposium Series, to be held Friday through Sunday, November 10-12, 1995, at 
the Massachusetts Institute of Technology.

The topics of the eight symposia in the 1995 Fall Symposium Series are:

-  Active Learning 
-  Adaptation of Knowledge for Reuse 
-  AI Applications in Knowledge Navigation and Retrieval 
-  Computational Models for Integrating Language and Vision 
-  Embodied Language and Action 
-  Formalizing Context 
-  Genetic Programming 
-  Rational Agency: Concepts, Theories, Models, and Applications

Symposia will be limited to between forty and sixty participants. Each
participant will be expected to attend a single symposium. Working notes 
will be prepared and distributed to participants in each symposium.

A general plenary session, in which the highlights of each symposium will be
presented, will be held on Saturday, November 11, and an informal reception 
will be held on Friday, November 10.

In addition to invited participants, a limited number of other interested
parties will be able to register in each symposium on a first-come, 
first-served basis. Registration will be available by 1 August, 1995. 
To obtain registration information write to the AAAI at 445 Burgess Drive, 
Menlo Park, CA 94025 (fss at aaai.org).

Submission Dates

-  Submissions for the symposia are due on April 14, 1995. 
-  Notification of acceptance will be given by May 19, 1995. 
-  Material to be included in the working notes of the symposium must be
   received by September 1, 1995.

See the appropriate section below for specific submission requirements for each
symposium.

This document is available as 
http://www.ai.mit.edu/people/las/aaai/fss-95/fss-95-cfp.html

*******************************************************************************

			ACTIVE LEARNING

An active learning system is one that can influence the training data it
receives by actions or queries to its environment. Properly selected, these
actions can drastically reduce the amount of data and computation required by a
machine learner. Active learning has been studied independently by researchers
in machine learning, neural networks, robotics, computational learning theory,
experiment design, information retrieval, and reinforcement learning, among
other areas. This symposium will bring researchers together to clarify the
foundations of active learning and point out synergies to build on.

Submission Information

Potential participants should submit a position paper (at most two pages)
discussing what the participant could contribute to a dialogue on active
learning and/or what they hope to learn by participating. Suggested topics
include:

Theory: What are the important results in the theory of active learning and what
are important open problems? How much guidance does theory give to application?

Algorithms: What successful algorithms have been found for active learning? How
general are they? For what tasks are they appropriate?

Evaluation: How can accuracy, convergence, and other properties of active
learning algorithms be evaluated when, for instance, data is not sampled
randomly?

Taxonomy: What kinds of information are available to learners (e.g. membership
vs. equivalence queries, labeled vs. unlabeled data) and what are the ways
learning methods can use them? What are the commonalities among methods studied
by different fields?

Papers should be sent to David D. Lewis, lewis at research.att.com, AT&T Bell
Laboratories, 600 Mountain Ave., Room 2C-408, Murray Hill, NJ 07974-0636.
Electronic mail submissions are strongly preferred.

Symposium Structure

The symposium will be broken into sessions, each dedicated to a major theme
identified within the position papers. Sessions will begin with a background
presentation by an invited speaker, followed by brief position statements from
selected participants. A significant portion of each session will be reserved
for group discussion, guided by a moderator and focused on the core issue for
the session. The final session of the symposium will accommodate new issues that
are raised during sessions.

Organizing Committee

David A. Cohn (cochair), MIT, cohn at psyche.mit.edu; David D. Lewis (cochair),
AT&T Bell Labs, lewis at research.att.com; Kathryn Chaloner, U. Minnesota ; Leslie
Pack Kaelbling, Brown U.; Robert Schapire, AT&T Bell Labs; Sebastian Thrun, U.
Bonn; Paul Utgoff, U. Mass Amherst.

******************************************************************************

		ADAPTATION OF KNOWLEDGE FOR REUSE

Several areas in AI address issues of creating and storing knowledge constructs
(such as cases, plans, designs, specifications, concepts, domain theories,
schedules). There is broad interest in reusing these constructs in similar
problem-solving situations so as to avoid expensive re-derivation. Adaptation
techniques have been developed to support reuse in frameworks such as analogical
problem solving, case-based reasoning, problem reformulation, or representation
change and task domains such as creativity, design, planning, program
transformation or software reuse, schedule revision, and theory revision.

However, many open issues remain, and progress on such issues as case adaptation
would substantially assist many researchers and practitioners. Our goals are to
characterize the approaches to adaptation employed in various AI subfields,
define the core issues in adaptation of knowledge, and advance the
state-of-the-art in addressing these issues. We intend that presentations will
investigate novel solutions to unsolved problems on adaptation, reflect diverse
viewpoints, and focus on adaptation issues that are common to several subfields
of AI. Discussions will be held on the strengths and limitations of adaptation
techniques and their interrelationships.

Invited talks will be given by experts who will discuss methods for the
adaptation of various types of knowledge constructs. Two panels will be held.
First, researchers studying knowledge adaptation from different perspectives
will discuss how approaches used in their community differ from those used
elsewhere, focusing on their potential benefits for other problems. Panelists in
the second panel will lead discussions on identifying the core issues in
knowledge adaptation raised in the presentations and the impact of the proposed
methods on addressing these issues.

Submission Information

Anyone interested in presenting relevant material is invited to email PostScript
submissions to aha at aic.nrl.navy.mil using the six-page AAAI-94 proceedings
format. Anyone interested in attending is asked to submit a two-page research
statement and a list of relevant publications. Please see
http://www.aic.nrl.navy.mil/~aha/aaai95-fss/home.html for further information.

Organizing Committee

David W. Aha (cochair), NRL, aha at aic. nrl.navy.mil; Brian Falkenhainer, Xerox;
Eric K. Jones, Victoria University; Subbarao Kambhampati, Arizona State
University; David Leake, Indiana University; Ashwin Ram (cochair), Georgia
Institute of Technology, ashwin at cc.gatech.edu.

*****************************************************************************

	AI APPLICATIONS IN KNOWLEDGE NAVIGATION AND RETRIEVAL

The diversity and volume of accessible on-line data is increasing dramatically.
As a result, existing tools for searching and browsing information are becoming
less effective. The increasing use of non-text data such as images, audio and
video has amplified this trend.

Knowledge navigation systems are knowledge-based interfaces to information
resources. They allow users to investigate the contents of complex and diverse
sources of data in a natural manner. For example, intelligent browsers that can
help direct a user through a large multi-dimensional information space, agents
that users can direct to perform information finding tasks, or knowledge-based
intermediaries that employ retrieval strategies to gather information relevant
to a userUs request.

The purpose of this symposium is to examine the state of the art in knowledge
navigation by examining existing applications and by discussing new techniques
and research directions. We encourage two types of submissions: work-in-progress
papers that point towards the future of this research area, and demonstrations
of knowledge navigation systems. Some research issues of interest:

-  Indexing: What indexing methods are appropriate and feasible for knowledge
   navigation systems? How can indices be extracted from data?

-  Retrieval: What retrieval methods are appropriate for knowledge navigation?
   What retrieval strategies can be employed?

-  Learning: How can knowledge navigation systems adapt to a changing knowledge
   environment and to user needs?

-  User interfaces: What are the characteristics of a useful navigational
   interface? What roles can or should an "agent" metaphor play in such 
   interfaces? How can a navigation system orient the user in the 
   information space?

-  Multi-source integration: How can multiple data and knowledge sources be
   integrated to address usersU needs?

-  Multimedia: What are the challenges presented by multimedia information
   sources?

Submission Information

The symposium will consist of invited talks, presentations, and hands-on
demonstration/discussion sessions. Interested participants should submit a short
paper (8 pages maximum) addressing a research issue in knowledge navigation or
describing a knowledge navigation system that can be made available for hands-on
demonstration at the symposium. System descriptions should clearly indicate the
novel and interesting features of the system to be presented and its
applicability to the central problems in knowledge navigation. Those wishing to
demonstrate should also include a one-page description of their hardware and
connectivity requirements. Send, by email,  either a URL pointing to a
PostScript version of the paper or the PostScript copy itself to 
aiakn at cs.uchicago.edu. Or, send 5 hard copies to Robin Burke, AI Applications 
in Knowledge Navigation, University of Chicago, Department of Computer Science.
1100 E. 58th St. Chicago, IL 60637. For further information, a web page for this
symposium is located at http://www-cs.uchicago.edu/ ~burke/aiakn.html

Organizing Committee

Robin Burke (chair), University of Chicago, burke at cs.uchicago.edu; Catharine
Baudin, NASA Ames; Su-Shing Chen, National Science Foundation; Kristian Hammond,
University of Chicago; Christopher Owens, Bolt, Beranek & Newman.

Program Committee

Ray Bariess, Institute for the Learning Sciences; Alon Levy, AT&T Bell
Laboratories; Jim Mayfield, University of Maryland, Baltimore County; Dick
Osgood, Andersen Consulting.

******************************************************************************

	COMPUTATIONAL MODELS FOR INTEGRATING LANGUAGE AND VISION

This symposium will focus on research issues in developing computational models
for integrating language and vision. The intrinsic difficulty of both natural
language processing and computer vision has discouraged researchers from
attempting integration, although in some cases it may simplify individual tasks
like collateral-based vision, resolving ambiguous sentences through the use of
visual information.

Developing a bridge between language and vision is nontrivial, because the
correspondence between words and images is not one-to-one. Much has been said
about the necessity of linking language and perception for a system to exhibit
intelligent behavior, but there has been relatively little work on developing
computational models for this task. A natural-language understanding system
should be able to understand and make references to the visual world. The use of
scene-specific context (obtained from written or spoken text accompanying a
scene) could greatly enhance the performance of computer vision systems.

Some topics to be addressed are:

-  use of collateral text in image and graphics understanding

-  generating natural-language descriptions of visual data (e.g., event
   perception in image sequences)

-  identifying and extracting visual information from language

-  understanding spatial language, spatial reasoning

-  knowledge representation for linguistic and visual information, hybrid
   (language and visual) knowledge bases

-  use of visual data in disambiguating/understanding text

-  content-based retrieval from integrated text/image databases

-  language-based scene modeling (e.g., picture or graphics generation)

-  cognitive theories connecting language and perception

Submission Information

The symposium will consist of invited talks, panel discussions, individual
presentations and group discussions. Those interested in making a presentation
should submit a technical paper (not to exceed 3,000 words). Other participants
should submit either a position paper or a research abstract. Email submissions
in postscript format are encouraged. Send to burhans @cs.buffalo.edu.
Alternatively, 4 hard copies may be sent to Rohini Srihari, CEDAR/SUNY at
Buffalo, UB Commons, 520 Lee Entrance Suite 202, Buffalo, NY 14228-2567. Further
information on this symposium may be found at
http://www.cedar.buffalo.edu/Piction/FSS95/CFP.html. Please address questions to
Debra Burhans (burhans at cs.buffalo.edu) or Rajiv Chopra (rchopra at cs.buffalo.edu).

Organizing Committee

Janice Glasgow, Queen's University; Ken Forbus, Northwestern University; Annette
Herskovits, Wellesley College; Gordon Novak, University of Texas at Austin;
Candace Sidner, Lotus Development Corporation; Jeffrey Siskind, University of
Toronto; Rohini K. Srihari (chair), CEDAR, SUNY at Buffalo,
rohini at cedar.buffalo. edu; Thomas M. Strat, SRI International; David Waltz, NEC
Research Institute.

*******************************************************************************

			EMBODIED LANGUAGE AND ACTION

This symposium focuses on agents that can use language or similar communication,
such as gesture, to facilitate extended interactions in a shared physical or
simulated world. We examine how this embodiment in a shared world both
stimulates communication and provides a resource for understanding it. Our focus
is on the design of artificial agents, implemented in software, hardware, or as
animated characters. Papers should clearly relate the technical content
presented to one of the following tasks:

-  Two or more communicating agents work together to construct, carry out
   maintenance on, or destroy a physical or simulated artifact (Collaborative
   Engagement)

-  An agent assists a human by fetching or delivering physical or software
   objects. The human communicates with the agent about what is to be fetched or
   delivered to where. (Delivery Assistance)

We solicit papers on the following issues (not to the exclusion of others):

-  Can task contexts act as resources for communication by simplifying the
   interpretation and production of communicative acts?

-  How does physical embodiment and its concomitant resource limitation affect
   an agentUs ability to interpret or generate language?

-  Can architectures designed to support perception and action support language
   or other forms of communication?

-  How can agents to mediate between the propositional representations of
   language and the (often) non-propositional representations of perception and
   action?

-  What tradeoffs exist between the use of communication to improve the agents'
   task performance and the additional overhead involved in understanding and
   generating messages?

-  Do differences between communication used to support concurrent task
   execution and communication used to support planning, reflect deeper 
   differences in agent ability?

-  What is the role of negotiation, whether of task responsibilities, or of
   reference and meaning, in such situated task environments?

Submission Information

Interested participants should submit either (1) a paper (in 12 pt font, not to
exceed 3000 words), or (2) a statement of interest briefly describing the
authorUs relevant work in this area and listing recent relevant publications.
Send contributions, plain ascii or postscript, to ian at ai. mit.edu. If electronic
submission is impossible, mail 6 copies to Ian Horswill, MIT Artificial
Intelligence Laboratory, 545 Technology Square, Cambridge, MA 02139.

Organizing Committee

John Batali, UCSD; Jim Firby, University of Chicago; Ian Horswill (cochair),
MIT, ian at ai.mit.edu; Marilyn Walker (cochair), Mitsubishi Cambridge Research
Labs, walker at merl.com; Bonnie Webber, University of Pennsylvania.

******************************************************************************

			FORMALIZING CONTEXT

The notion of context has played an important role in AI systems for many years.
However, formal logical explication of contexts remains an area of research with
significant open issues. This symposium will provide a forum for discussing
formalizations of contexts, approaches to resolving open issues, and application
areas for context formalisms.

The most ambitious goal of formalizing contexts is to make automated reasoning
systems which are never permanently stuck with the concepts they use at a given
time because they can always transcend the context they are in. Such a
capability would allow the designer of a reasoning system to include only such
phenomena as are required for the system's immediate purpose, retaining the
assurance that if a broader system is required later, "lifting rules" can be
devised to restate the facts from the narrow context in the broader context with
qualifications added as necessary. A formal theory of context in which sentences
are always considered as asserted within a context could provide a basis for
such transcendence.

Formal theories of context are also needed to provide a representation of the
context associated with a particular circumstance, e.g. the context of a
conversation in which terms have particular meanings that they wouldn't have in
the language in general. Linguists and philosophers have already studied similar
notions of context. An example is the situation theory that has been proposed in
philosophy and applied to linguistics. However, these theories usually lie
embedded in the analysis of specific linguistic constructions, so locating the
exact match with AI concerns is itself a research challenge.

This symposium aims to bring together researchers who have studied or applied
contexts in AI or related fields. Technical papers dealing with formalizations
of context, the problem of generality, and use of context in common sense
reasoning are especially welcome. However, survey papers which focus on contexts
from other points of view, such as philosophy, linguistics, or natural language
processing, or which apply contexts in other areas of AI, are also encouraged.

Submission Information

Persons wishing to make presentations should submit papers (up to 12 pages, 12
pt font). Persons wishing only to attend should submit a 1-2 page research
summary including a list of relevant publications. A PostScript file or 8 paper
copies should be sent to the program chair, Sasa Buvac Department of Computer
Science Stanford University Stanford CA 94305-2140 buvac at sail.stanford.edu.
Limited funding will be available to support student travel.

Organizing Committee

Sasa Buvac (chair), Stanford University, buvac at sail.stanford.edu; Richard Fikes,
Stanford University; Ramanathan Guha, MCC; Pat Hayes, Beckman Institute; John
McCarthy, Stanford University; Murray Shanahan, Imperial College; Robert
Stalnaker, MIT; Johan van Benthem, University of Amsterdam. Genetic programming
(GP) extends the genetic algorithm to the domain of computer programs. In
genetic programming, populations of programs are genetically bred to solve
problems. Genetic programming can solve problems of system identification,
classification, control, robotics, optimization, game-playing, and pattern
recognition.

******************************************************************************

			GENETIC PROGRAMMING

Starting with a primordial ooze of hundreds or thousands of randomly created
programs composed of functions and terminals appropriate to the problem, the
population is progressively evolved over a series of generations by applying the
operations of Darwinian fitness proportionate reproduction and crossover (sexual
recombination).   
Topics of interest for the symposium include:
 
-  The theoretical basis of genetic programming 
-  Applications of genetic programming
-  Rigorousness of validation techniques 
-  Hierarchical decomposition, e.g. automatically defined functions 
-  Competitive coevolution 
-  Automatic parameter tuning 
-  Representation issues 
-  Genetic operators 
-  Establishing standard benchmark problems 
-  Parallelization techniques 
-  Innovative variations 

Submission Information

The format of the symposium will encourage interaction and discussion, but 
will also include formal presentations. Persons wishing to make a presentation 
should submit an extended abstract of up to 2500 words of their work in 
progress or completed work. For accepted abstracts, full papers will be due 
at a date closer to the symposium.

Persons not wishing to make a presentation are asked to submit a one-page 
description of their research interests since there may be limited room 
for participation.

Submit your abstract or one-page description as plain text electronically 
by Friday April 14, 1995, with a hard-copy backup to Eric V. Siegel, AAAI 
GP Symposium Co-Chair, Columbia University, Department of Computer Science,
500 W 120th Street, New York, NY 10027, USA; telephone: 212-939-7112, 
fax: 212-666-0140, e-mail: evs at cs.columbia.edu.

Organizing Committee

Robert Collins, USAnimation, Inc.; Frederic Gruau, Stanford University; 
John R. Koza (co-chair), Stanford University, koza at cs.stanford.edu; 
Robert Collins, US Animation, Inc.; Conor Ryan, University College Cork; 
Eric V. Siegel (co-chair), Columbia University, evs at cs.columbia.edu; 
Andy Singleton, Creation Mechanics, Inc.; Astro Teller, Carnegie-Mellon 
University.

******************************************************************************

RATIONAL AGENCY: CONCEPTS, THEORIES, MODELS, AND APPLICATIONS

This symposium explores conceptions of rational agency and their implications
for theory, research, and practice. The view that intelligent systems are, or
ought to be, rational agents underlies much of the theory and research in
artificial intelligence and cognitive science. However, no consensus exists on a
proper view of agency or rationality principles for practical agents.

Traditionally agents are presumed disposed toward purposive action. However,
agent theories abound in which behavior is fundamentally reactive. Some theories
emphasize agents' abilities to manage private belief systems. Others focus on
agents' interactions with their environment, sometimes including other agents.
Application builders have recently broadened the term "agent" to mean any
embedded system performing tasks to support human users.

Rationality accounts are equally diverse. Rationality involves having reasons
warranting particular beliefs (epistemic rationality) or particular desires and
actions (strategic or practical rationality). Many agent models propose
epistemic rationality criteria such as logical consistency or consequential
closure. Other agent models base practical rationality on classical or
non-monotonic logics for reasoning about action. Such logicist views are now
being challenged by decision theoretic accounts emphasizing optimal action under
uncertainty, including recent work on decision theoretic principles of
metareasoning for limited rationality.

Our symposium will explore the diverse views of rational agency. Through
informal presentations and group discussion, participants will critically
examine agency concepts and rationality principles, review computational agent
models and applications, and consider promising directions for future work on
this topic.

Submission Information

Prospective participants should submit a brief paper (5 pages or less)
describing their research in relation to any of the following questions:

-  Is rationality important; must an agent be rational to be successful?

-  What are suitable principles of epistemic, strategic, and limited
   rationality?

-  Are rationality principles applicable to retrospective processes such as
   learning?

-  What are general requirements on rational agent architectures?

-  How, if at all, must a model of rational agency be modified to account for
   social, multi-agent interaction?

Those wishing to make a specific presentation should describe its contents in
their concept paper. Note: While we recognize that our topic lends itself to
formal analysis, we also encourage discussion of experimental work with
implemented agents. Postscript files of concept papers should be sent by email
only to the program chair, fehling at lis.stanford.edu.

Organizing Committee

Michael Fehling (chair), Stanford University, fehling at lis.stanford.edu; Don
Perlis, University of Maryland; Martha Pollack, University of Pittsburgh; John
Pollock, University of Arizona.






From tesauro at watson.ibm.com  Thu Dec  8 14:28:41 1994
From: tesauro at watson.ibm.com (tesauro@watson.ibm.com)
Date: Thu, 8 Dec 94 14:28:41 EST
Subject: NIPS*94 Top Ten List
Message-ID: <mailman.715.1149540302.24850.connectionists@cs.cmu.edu>

By popular demand, here once again is the Top Ten List.
Thanks go to my comedy-writing collaborators, Mike Mozer
and Don Mathis of the Univ. of Colorado.
--Gerry Tesauro
------------------------------------------------------------------
         Top 10 Little-Known Improvements to NIPS This Year

10.  No more of that annoying math!

 9.  To keep speakers on their toes, a gong at every table.

 8.  Special prize for "Most Laughable Poster"

 7.  Push pins available to everyone, not just to poster presenters.

 6.  Special evening session in which senior researchers
     display their surgical scars.

 5.  Hotel staff all licensed to back-propagate.

 4.  More rock, less talk.

 3.  Student volunteers at the registration desk will finally
     stop asking "You want fries with that?"

 2.  Invited speakers disqualified if they test positive
     for steroids.

 1.  Morgan Kaufmann to publish special "Swimsuit Edition"
     of conference proceedings.


From dwang at cis.ohio-state.edu  Mon Dec 12 10:50:23 1994
From: dwang at cis.ohio-state.edu (DeLiang Wang)
Date: Mon, 12 Dec 1994 10:50:23 -0500
Subject: About sequential learning (or interference)
Message-ID: <mailman.716.1149540302.24850.connectionists@cs.cmu.edu>

It was reported some time ago that multilayer perceptrons suffer the problem
of so-called "catastrophic interference", meaning that later training will
destroy previously acquired "knowledge" (see McCloskey & Cohen, Psychol. of
Learning and Motivat. 24, 1990; Ratcliff, Psychol. Rev. 97, 1990). This seems
to be a serious problem, if we want to use neural networks both as a stable
knowledge store and a long-term problem solver.

The problems seems to exist in associative memory models as well, even though
the simple prescription of the Hopfield net can easily incorporate more
patterns as long as they are within the capacity limit. But we all know that
the original Hopfield net does not work well for correlated patterns, which
are the most interesting ones for real applications. Kanter and Sompolinsky
proposed a prescription to tackle the problem (Phys. Rev. A, 1987). 
Their prescription, however, requires nonlocal learning. 

Diederich and Opper (1987, Phys. Rev. Lett.) later proposed a local, 
iterative learning rule (similar to perceptron learning) that they show will 
converge to the prescription (which is the old idea of orthogonalization). 
According to their paper, to learn a new pattern one needs to bring in all 
previously acquired patterns during iterative training in order 
to make all of the patterns converge to the desired prescription.  Because
of this learning scheme, I suspect that the algorithm of Diederich and Opper
also suffers "catastrophic forgetting".

Is that a fair assessment of the problem?  Has any major effort been taken to 
address this important problem? 

DeLiang Wang


From jaap.murre at mrc-apu.cam.ac.uk  Tue Dec 13 11:59:33 1994
From: jaap.murre at mrc-apu.cam.ac.uk (Jaap Murre)
Date: Tue, 13 Dec 94 16:59:33 GMT
Subject: hypertransfer and interference
Message-ID: <9412131659.AA03848@rigel.mrc-apu.cam.ac.uk>

DeLiang Wang asks whether any major effort has been undertaken to
investigate and possibly remedy the problem of catastrophic interference.

A fairly detailed analysis of this problem can be found in:

Murre, J.M.J. (1992b). Categorization and learning in modular neural
      networks. Hemel Hempstead: Harvester Wheatsheaf. Co-published
      by Lawrence Erlbaum in the USA and Canada (Hillsdale, NJ).

I have recently completed a paper that shows that backpropagation not
only suffers from 'catastrophic interference' but also from 'hyper
transfer', i.e., in some circumstances performance on a set A actually
*improves* when learning a second set B. The learning transfer effects
are catastrophic (or hyper) with respect to human learning data. The
paper also shows that two-layer networks do not suffer from excessive
transfer and are in fact in very close accordance with the human
interference data as summarized in the classic paper by Osgood (1949):

Murre, J.M.J. (in press). Transfer of learning in backpropagation
      networks and in related neural network models. To appear in
      Levy, Bairaktaris, Bullinaria, & Cairns (Eds.), Connectionist
      Models of Memory and Language. London: UCL Press. (in fpt
      directory: hyper1.ps)

I have put the paper in our anonymous ftp directory:

      ftp ftp.mrc-apu.cam.ac.uk
      cd /pub/nn/murre
      bin
      get hyper1.ps



From lpratt at franklinite.Mines.Colorado.EDU  Tue Dec 13 05:20:35 1994
From: lpratt at franklinite.Mines.Colorado.EDU (Lorien Y. Pratt)
Date: Tue, 13 Dec 94 11:20:35 +0100
Subject: About sequential learning (or interference)
Message-ID: <9412131020.AA01764@franklinite.Mines.Colorado.EDU>

Deliang,

My own work on transfer between neural networks has addressed the sequential 
learning problem, see pratt-93, cited below, and lots of other papers available 
from my web page: http://vita.mines.colorado.edu:3857/1s/lpratt.  My formuation 
differs from others in that, if the new task is essentially different from the 
old task, I intentionally lose old-task performance if necessary to improve on 
the new task.  Other people who have looked at a similar formulation include 
sharkey-92, sharkey-93, naik-92b, and agarwal-92, martin-88.  Those who have 
attempted to preserve old-task performance include mccloskey-89, and recent 
work by Thrun & Mitchell, especially thrun-93a, but also including thrun-92, 
thrun-93b. Sebastian and I had a long talk at NIPS about a general formulation 
for transfer as a solution to the sequential learning problem -- he's got an 
excellent formulation along these lines.

Approaches to modular training can also be viewed as one approach to handling 
the sequential learning problem, in the special case where the old tasks can be 
viewed as subtasks of the new task (i.e. pratt-91, waibel-89, and Jacobs' and 
others' recent work on mixtures of experts, including Pomerleau's approach to 
Alvinn, see last year's NIPS).

My thesis (see my web page) has a more detailed review of all this stuff.

Hope that helps.
  --Lori, recovering from NIPS
  

  @incollection{ pratt-93,
MYKEY           = " pratt-93 : .con .bap",
EDITOR          = "C.L. Giles and S. J. Hanson and J. D. Cowan",
BOOKTITLE       = "{Advances in Neural Information Processing Systems 5}",
AUTHOR          = "L. Y. Pratt",
TITLE           = "Discriminability-Based Transfer between Neural Networks",
ADDRESS         = "San Mateo, CA",
PUBLISHER       = "Morgan Kaufmann Publishers",
YEAR            = 1993,
PAGES           = {204--211},
CATALOGDATE     = "April 12, 1993",
NOTE            = "Also available via anonymous ftp to
franklinite.mines.colorado.edu: pub/pratt-papers/pratt-nips5.ps.Z",
}
@inproceedings{ sharkey-92,
MYKEY           = " sharkey-92 : ",
TITLE           = "Adaptive generalisation and the transfer of knowledge",
AUTHOR          = "Noel E. Sharkey and Amanda J. C. Sharkey",
BOOKTITLE       = "Proceedings of the Second Irish Neural Networks Conference,
                   Belfast",
YEAR            = 1992,
CATALOGDATE     = "December 18, 1992",
}

@article{ sharkey-93,
MYKEY           = " sharkey-93 : ",
TITLE           = "Adaptive generalisation and the transfer of knowledge",
AUTHOR          = "N. E. Sharkey and A. J. C. Sharkey",
NUMBER          = "In press",
VOLUME          = "Special Issue on Connectionism",
JOURNAL         = "AI Review",
YEAR            = 1993,
ANNOTE          = "In press",
CATALOGDATE     = "May 19, 1993",
}
@inproceedings{ naik-92b,
MYKEY           = " naik-92b : .eml .unb .con ",
AUTHOR          = "D. K. Naik and R. J. Mammone and A. Agarwal",
TITLE           = "Meta-Neural Network approach to learning by learning",
YEAR            = 1992,
BOOKTITLE       = "Intelligence Engineering Systems through Artificial Neural
                   Networks",
ORGANIZATION    = "The American Society of Mechanical Engineers",
PUBLISHER       = "ASME Press",
VOLUME          = 2,
PAGES           = {245--252},
CATALOGDATE     = "December 23, 1992",
}

@inproceedings{ agarwal-92,
MYKEY           = " agarwal-92 : .eml .unb .con ",
AUTHOR          = "A. Agarwal and R. J. Mammone and D. K. Naik",
TITLE           = "An on-line Training Algorithm to Overcome Catastrophic
                   Forgetting",
BOOKTITLE       = "Intelligence Engineering Systems through Artificial Neural
                   Networks",
YEAR            = 1992,
ORGANIZATION    = "The American Society of Mechanical Engineers",
PUBLISHER       = "ASME Press",
VOLUME          = 2,
PAGES           = {239--244},
CATALOGDATE     = "December 23, 1992",
}

@techreport{ martin-88,
MYKEY           = " martin-88 : ",
TITLE           = "The Effects of Old Learning on New in Hopfield and
                   Backpropagation Nets",
AUTHOR          = "Gale Martin",
KEY             = "martin-88",
INSTITUTION     = "Microelectronics and Computer Technology Corporation (MCC)",
NUMBER          = "ACA-HI-019",
YEAR            = 1988,
CATALOGDATE     = "January 3, 1992",
}

@article{ mccloskey-89,
MYKEY           = " mccloskey-89 : .unb .con .csy .adap",
TITLE           = "Catastrophic interference in connectionist networks: the
                   sequential learning problem",
KEY             = "mccloskey-89",
AUTHOR          = "Michael McCloskey and Neal J. Cohen",
JOURNAL         = "The psychology of learning and motivation",
VOLUME          = 24,
YEAR            = 1989,
CATALOGDATE     = "April 9, 1991",
}
@inproceedings{ thrun-93a,
MYKEY           = " thrun-93a : ",
TITLE           = "Lifelong Robot Learning",
AUTHOR          = "Sebastian B. Thrun and Tom M. Mitchell",
BOOKTITLE       = "Proceedings of the {NATO} {ASI}: The biology and technology
                   of intelligent autonomous agents",
EDITOR          = "Luc Steels",
YEAR            = 1993,
KEY             = "thrun-93a",
CATALOGDATE     = "September 1, 1993",
}

@INPROCEEDINGS{thrun-93b,
AUTHOR        = {Thrun, Sebastian B. and Mitchell, Tom M.},
TITLE         = {Integrating Inductive Neural Network Learning and
                 Explanation-Based Learning},
BOOKTITLE     = {Proceedings of IJCAI-93},
YEAR          = {1993},
ORGANIZATION  = {IJCAI, Inc.},
PUBLISHER     = {},
ADDRESS       = {Chamberry, France},
MONTH         = {},
CATALOGDATE     = "October 11, 1993",
}

@inproceedings{ pratt-91,
MYKEY           = " pratt-91 : .min .bap .app .spc .con ",
AUTHOR          = "Lorien Y. Pratt and Jack Mostow and Candace A. Kamm",
TITLE           = "Direct Transfer of Learned Information among Neural
                    Networks",
BOOKTITLE       = "Proceedings of the Ninth National Conference on
                   Artificial Intelligence (AAAI-91)",
PAGES           = {584--589},
ADDRESS         = "Anaheim, CA",
YEAR            = 1991,
}

@article{ waibel-89,
MYKEY           = " waibel-89 : .bap .unr .unb .tem .spc .con ",
TITLE           = "Modular Construction of Time-Delay Neural Networks for
                    Speech Recognition",
AUTHOR          = "Alexander Waibel",
journal         = "Neural Computation",
volume          = 1,
pages           = {39--46},
year            = 1989
}



From wermter at nats2.informatik.uni-hamburg.de  Tue Dec 13 06:04:05 1994
From: wermter at nats2.informatik.uni-hamburg.de (Stefan Wermter)
Date: Tue, 13 Dec 94 12:04:05 +0100
Subject: book on hybrid connectionist language processing
Message-ID: <9412131104.AA01259@nats2.informatik.uni-hamburg.de>



BOOK ANNOUNCEMENT
-----------------

The following book is now available from the beginning of December 1994.





Title:      Hybrid connectionist natural language processing

Date:       1995

Author:     Stefan Wermter
            Dept. of Computer Science
            University of Hamburg
            Vogt-Koelln-Str. 30
            D-22527 Hamburg
            Germany

            wermter at informatik.uni-hamburg.de

Series:     Neural Computing Series 7

Publisher:  Chapman & Hall Inc
            2-6 Boundary Row
            London SE1 8HN
            England



(Order information in the end of this message)



Description
-----------

The objective of this  book is  to describe a  new approach  in hybrid
connectionist natural language processing which bridges the gap between
strictly symbolic and connectionist systems. This objective is tackled
in two ways: the book gives an overview of hybrid connectionist archi-
tectures for natural  language processing; and  it demonstrates that a
hybrid connectionist architecture can be used for learning real-world
natural language problems. The book is primarily intended for scientists
and students interested in the fields of artificial intelligence, neural
networks,  connectionism, natural  language processing, hybrid symbolic
connectionist architectures,  parallel distributed processing, machine
learning, automatic knowledge  acquisition or computational linguistics.
Furthermore,  it might  be of  interest for  scientists and students
in information retrieval and cognitive science, since the book points
out interdisciplinary relationships to these fields.

We develop a systematic spectrum of hybrid connectionist architectures,
from completely symbolic architectures to separated hybrid connectionist
architectures,  integrated  hybrid  connectionist architectures and
completely connectionist architectures. Within this systematic spectrum
we have designed a system SCAN with two separated hybrid  connectionist
architectures and two integrated hybrid connectionist architectures for
a scanning  understanding of phrases. A scanning understanding  is a
relation-based flat understanding in  contrast to traditional symbolic
in-depth understanding. Hybrid  connectionist representations  consist
of either a combination of connectionist and symbolic representations
or different  connectionist representations. In particular, we focus on
important  tasks like structural  disambiguation and semantic context
classification. We show that  a parallel modular, constraint-based,
plausibility-based  and learned use of multiple hybrid connectionist
representations provides powerful architectures  for learning a scanning
understanding.  In particular, the combination of direct encoding of
domain-independent structural knowledge and the connectionist learning of
domain-dependent semantic knowledge, as suggested by a scanning under-
standing in SCAN, provides concepts which lead to  flexible, adaptable,
transportable architectures for different domains.




Table of Contents
-----------------

1 Introduction
  1.1 Learning a Scanning Understanding
  1.2 The General Approach
  1.3 Towards a Hybrid Connectionist Memory Organization
  1.4 An Overview of the SCAN Architecture
  1.5 Organization and Reader's Guide

2 Connectionist and Hybrid Models for Language Understanding
  2.1 Foundations of Connectionist and Hybrid Connectionist Approaches
  2.2 Connectionist Architectures
    2.2.1 Representation of Language in Parallel Spatial Models
      Early Pattern Associator for Past Tense Learning
      Pattern Associator for Semantic Case Assignment
      Pattern Associator with Sliding Window
      Time Delay Neural Networks
    2.2.2 Representation of Language in Recurrent Models
      Recurrent Jordan Network for Action Generation
      Simple Recurrent Network for Sequence Processing
      Recursive Autoassociative Memory Network
    2.2.3 Towards Modular and Integrated Connectionist Models
      Cascaded Networks
      Sentence Gestalt Model
      Grounding Models
  2.3 Hybrid Connectionist Architectures
    2.3.1 Sentence Analysis in Hybrid Models
      Hybrid Interactive Model for Constraint Integration
      Hybrid Model for Sentence Analysis
    2.3.2 Inferencing in Hybrid Models
      Symbolic Marker Passing and Localist Networks
      Symbolic Reasoning with Connectionist Models
    2.3.3 Architectural Issues in Hybrid Connectionist Systems
      Symbolic Neuroengineering and Symbolic Recirculation
      Modular Model for Parsing
  2.4 Summary and Discussion

3 A Hybrid Connectionist Scanning Understanding of Phrases
  3.1 Foundations of a Hybrid Connectionist Architecture
    3.1.1 Motivation for a Hybrid Connectionist Architecture
    3.1.2 The Computational Theory Level for a Scanning Understanding
    3.1.3 Constraint Integration
    3.1.4 Plausibility view
    3.1.5 Learning
    3.1.6 Subtasks of Scanning Understanding at the Computational Theory Level
    3.1.7 The Representation Level for a Scanning Understanding
  3.2 Corpora and Lexicon for a Scanning Understanding
    3.2.1 The Underlying Corpora
    3.2.2 Complex Phrases
    3.2.3 Context and Ambiguities of Phrases
    3.2.4 Organization of the Lexicon
  3.3 Plausibility Networks
    3.3.1 Learning Semantic Relationships and Semantic Context
    3.3.2 The Foundation of Plausibility Networks
    3.3.3 Plausibility Networks for Noun-Connecting Semantic Relationships
    3.3.4 Learning in Plausibility Networks
    3.3.5 Recurrent Plausibility Networks for Contextual Relationships
    3.3.6 Learning in Recurrent Plausibility Networks
  3.4 Summary and Discussion

4 Structural Phrase Analysis in a Hybrid Separated Model
  4.1 Introduction and Overview
  4.2 Constraints for Coordination
  4.3 Symbolic Representation of Syntactic Constraints
    4.3.1 A Grammar for Complex Noun Phrases
    4.3.2 The Active Chart Parser and the Syntactic Constraints
  4.4 Connectionist Representation of Semantic Constraints
    4.4.1 Head-noun Structure for Semantic Relationships
    4.4.2 Training and Testing Plausibility Networks with NCN-relationships
    4.4.3 Learned Internal Representation
  4.5 Combining Chart Parser and Plausibility Networks
  4.6 A Case Study
  4.7 Summary and Discussion

5 Structural Phrase Analysis in a Hybrid Integrated Model
  5.1 Introduction and Overview
  5.2 Constraints for Prepositional Phrase Attachment
  5.3 Representation of Constraints in Relaxation Networks
    5.3.1 Integrated Relaxation Network
    5.3.2 The Relaxation Algorithm
    5.3.3 Testing Relaxation Networks
  5.4 Representation of Semantic Constraints in Plausibility Networks
    5.4.1 Training and Testing Plausibility Networks with NPN-Relationships
    5.4.2 Learned Internal Representation
  5.5 Combining Relaxation Networks and Plausibility Networks
    5.5.1 The Interface between Relaxation Networks and Plausibility Networks
    5.5.2 The Dynamics of Processing in a Relaxation Network
  5.6 A Case Study
  5.7 Summary and Discussion

6 Contextual Phrase Analysis in a Hybrid Separated Model
  6.1 Introduction and Overview
  6.2 Towards a Scanning Understanding of Semantic Phrase Context
    6.2.1 Superficial Classification in Information Retrieval
    6.2.2 Skimming Classification with Symbolic Matching
  6.3 Constraints for Semantic Context Classification of Noun Phrases
  6.4 Syntactic Condensation of Phrases to Compound Nouns
    6.4.1 Motivation of Symbolic Condensation
    6.4.2 Condensation Using a Symbolic Chart Parser
  6.5 Plausibility Networks for Context Classification of Compound Nouns
    6.5.1 Training and Testing the Recurrent Plausibility Network
    6.5.2 Learned Internal Representation
  6.6 Summary and Discussion

7 Contextual Phrase Analysis in a Hybrid Integrated Model
  7.1 Introduction and Overview
  7.2 Constraints for Semantic Context Classification of Phrases
  7.3 Plausibility Networks for Context Classification of Phrases
    7.3.1 Training and Testing with Complete Phrases
    7.3.2 Training and Testing with Phrases without Insignificant Words
    7.3.3 Learned Internal Representation
  7.4 Semantic Context Classification and Text Filtering
  7.5 Summary and Discussion

8 General Summary and Discussion
  8.1 The General Framework of SCAN
  8.2 Analysis and Evaluation
    8.2.1 Evaluating the Problems
    8.2.2 Evaluating the Methods
    8.2.3 Evaluating the Representations
    8.2.4 Evaluating the Experiment Design
    8.2.5 Evaluating the Experiment Results
  8.3 Extensions of a Scanning Understanding
    8.3.1 Extending Modular Subtasks
    8.3.2 Extending Interactions
  8.4 Contributions and Conclusions

9 Appendix
  9.1 Hierarchical Cluster Analysis
  9.2 Implementation
  9.3 Examples of Phrases for Structural Phrase Analysis
  9.4 Examples of Phrases for Contextual Phrase Analysis

References
Index







Orders information
-----------------

            ISBN: 0 412 59100 6
            Pages: 190
            Figures: 56

            Price: 29.95 pounds sterling, 52.00 US dollars
            Credit cards: all major credit cards accepted by Chapman & Hall


            Please order from:

        --  Pam Hounsome
            Chapman & Hall
            Cheriton House
            North Way
            Andover
            Hants, SP10 5BE, UK
            England
 
            UK orders:
               Tel: 01264 342923 Fax: 01264 364418
            Overseas orders:
               Tel: +44 1264 342830 Fax: +44 1264 342761
 

            (Sister company in US)
        --  Peter Clifford
            Chapman & Hall Inc 
            One Penn Plaza
            41st Floor
            New York NY 10119
            USA
 
            Tel: 212 564 1060
            Fax: 212 564 1505

 
        --  or e-mail on: order at Chaphall.com




















From mario at physics.uottawa.ca  Tue Dec 13 09:01:44 1994
From: mario at physics.uottawa.ca (Mario Marchand)
Date: Tue, 13 Dec 94 10:01:44 AST
Subject: Neural net and PAC learning paper: errata
Message-ID: <9412131401.AA27027@physics.uottawa.ca>

About a week ago, I have sent you this message:


>The following paper, which was presented at the NIPS'94 conference,
>is available by anonymous ftp at:
>
>ftp://dirac.physics.uottawa.ca/usr2/ftp/pub/tr/marchand
>
>
>FileName: nips94.ps
>
>Title: Learning Stochastic Perceptrons Under k-Blocking Distributions
>
>Authors: Marchand M. and Hadjifaradji S.
>
>
>ALSO: you will find other papers co-authored by Mario Marchand in
>      this directory. The text file: Abstracts-mm.txt contains a
>      list of abstracts of all the papers.


In fact the right path for ftp is:

ftp://dirac.physics.uottawa.ca/pub/tr/marchand

and the file name is: nips94.ps.Z    (compressed)

sorry for this inconvenience and many thanks for those who have
kindly reported the mistakes.

- mario





   ----------------------------------------------------------------
  |  UUU            UUU            Mario Marchand                  |
  |  UUU            UUU            -----------------------------   |
  |  UUU       OOOOOOOOOOOOOOOO    Department of Physics           |
  |  UUU      OOO   UUU      OOO   University of Ottawa            |
  |   UUUUUUUUUUUUUUUU       OOO   150 Louis Pasteur street        |
  |           OOO            OOO   PO BOX 450 STN A                |
  |            OOOOOOOOOOOOOOOO    Ottawa (Ont) Canada K1N 6N5     |
  |                                                                |
  |   ***** Internet E-Mail: mario at physics.uottawa.ca **********   |
  |   ***** Tel: (613)564-9293 ------------- Fax: 564-6712 *****   |
   ----------------------------------------------------------------





From bisant at gl.umbc.edu  Tue Dec 13 20:21:48 1994
From: bisant at gl.umbc.edu (Mr. David Bisant)
Date: Tue, 13 Dec 1994 20:21:48 -0500
Subject: FLAME: WCNN'95 - science in the mud pits
In-Reply-To: <199412081651.LAA01963@garnet.cs.brandeis.edu>
Message-ID: <Pine.SGI.3.90.941213195922.11172B-100000@umbc8.umbc.edu>

On Thu, 8 Dec 1994, Jordan Pollack wrote:

> > 
> >Authors must submit 
> >registration payment with papers to be eligible for the early 
> >registration fee...Registration fee includes 1995 membership and a 
> >one (1) year subscription to  the Journal Neural Networks.
> >A $35 per paper processing fee must be enclosed 
> >in order for the paper to be refereed.
> 
> Reading fees are charged by agents for aspiring science fiction
> writers. Linking registration to submission is nothing more than an
> admission that all papers are to be accepted.  And a fee of $480 is
> paying a lot more than membership, subscription, and a few days of
> coffee breaks!
> 

The new fee structure for WCNN did not strike me as being unusually
high.  I also think it will help solve a problem I observed last 
year with a few authors not showing up for their talks, which wasted
valuable time slots for oral presentation and upset schedules.  

With interest in the neural network field diminishing, new problems
in conferencing appear.  I think their fee structure is a reasonable 
approach to solving these problems.

David Bisant
(no affiliation with the conference organization)


From cga at cc.gatech.edu  Tue Dec 13 21:35:22 1994
From: cga at cc.gatech.edu (Christopher G. Atkeson)
Date: Tue, 13 Dec 1994 21:35:22 -0500
Subject: About sequential learning (or interference)
Message-ID: <199412140235.VAA17037@lennon.cc.gatech.edu>

Memory-Based Learning is one approach to avoiding interference,
and is in part descended from nearest neighbor neural networks, or at
least neural networks with local representations.

Some of our references that will point you to other work as well:

Atkeson, C.G. (1992) "Memory-Based Approaches To Approximating
Continuous Functions." In: Casdagli, M., & Eubank, S. (eds.),
"Nonlinear Modeling and Forecasting" Addison Wesley, Redwood City, CA.

Atkeson, C.G. ``Using Local Models to Control Movement'',
Proceedings, Neural Information Processing Systems,
Dec 1989, Denver, Colorado.
In: Neural Information Processing Systems 2.
Morgan Kaufmann, 1990.

Schaal, S., & Atkeson, C.G. (1994).
"Robot Juggling: An Implementation of Memory-Based Learning."
IEEE Control Systems Magazine, vol. 15, no. 1, pp. 57-71.

Chris Atkeson
Stefan Schaal


From jlm at crab.psy.cmu.edu  Tue Dec 13 23:16:34 1994
From: jlm at crab.psy.cmu.edu (James L. McClelland)
Date: Tue, 13 Dec 94 23:16:34 EST
Subject: About sequential learning (or interference)
In-Reply-To: <Added.8ivEqJi00UdbIelU4r@andrew.cmu.edu> (message from DeLiang Wang on Mon, 12 Dec 1994 10:50:23 -0500)
Message-ID: <9412140416.AA17263@crab.psy.cmu.edu.psy.cmu.edu>


DeLiang Wang <dwang at cis.ohio-state.edu> writes:

> It was reported some time ago that multilayer perceptrons suffer the problem
> of so-called "catastrophic interference", meaning that later training will
> destroy previously acquired "knowledge" (see McCloskey & Cohen, Psychol. of
> Learning and Motivat. 24, 1990; Ratcliff, Psychol. Rev. 97, 1990). This seems
> to be a serious problem, if we want to use neural networks both as a stable
> knowledge store and a long-term problem solver.

The brain appears to have solved this problem by storing new
information in a special associative memory system in the hippocampus.
According to this view, cortical (and some other non-hippocampal)
systems learn slowly, using what I call 'interleaved learning'.
Weights are adjusted a small amount after each experience, so that the
overall direction of weight change is governed by the structure
present in the ensemble of events and experiences.  New material can
be added to such a memory without catastrophic intereference if it is
added slowly, interleaved with ongoing exposure to other events and
experiences.  This is too slow for the demands placed on memory by the
world.  To allow rapid learning of new material without catastrophic
interference, new material is initially stored in the hippocampus,
where sparse, conjunctive representations are used to minimize
interference with other memories.  Reinstatement of these patterns
occurs in task-relevant situations via bi-directional connections
between hippocampus and cortex together with pattern completion within
the hippocampal formation.  Interestingly, it appears that
reinstatement occurs in off-line situations as well -- most notably,
during sleep.  Each reinstatement provides an opportunity for the
neocortex to learn; but these reinstatements occur interleaved with
other ongoing events and experiences, and weight changes are small on
each reinstatement, so that the neocortex learns the new material via
interleaved learning.  This theory is consistent with a lot of data at
this point, including of course the basic fact that bilateral removal
of the hippocampus leads to a profound deficit in the ability to form
arbitrary new memories rapidly.  Among the most important additional
findings are 1) the phenomenon of temporally graded retrograde amnesia
--- the finding that damage to the hippocampus can produce a selective
deficit for recently acquired memories leaving memories acquired
several weeks or months earlier intact --- and 2) the finding that
neurons in the hippocampus that are coactive during a particular
behavior appear to be coactive during sleep following the behavior, as
if the patterns of activity that were present during behavior are
reactivated during sleep.

I announced a technical report discussing the above ideas last March.
In that report, we focused on why it makes sense from a connectionist
point of view that the system should be organized as described above.
A reprint of the announcement (sans abstract) appears below.  This TR
is currently under revision for publication; the revised version will
contain a fuller presentation of the physiological evidence than the
present version, and I will announce it on connectionists when it
becomes available.

========================================================================
Technical report announcement:
------------------------------------------------------------------------

             Why there are Complementary Learning Systems
                  in the Hippocampus and Neocortex:
             Insights from the Successes and Failures of
             Connectionist Models of Learning and Memory

    James L. McClelland, Bruce L. McNaughton & Randall C. O'Reilly
        Carnegie Mellon University & The University of Arizona

                    Technical Report PDP.CNS.94.1
                             March, 1994

=======================================================================

Retrieval information:

unix> ftp 128.2.248.152                 # hydra.psy.cmu.edu
Name: anonymous
Password: <email address>
ftp> cd pub/pdp.cns
ftp> binary
ftp> get pdp.cns.94.1.ps.Z
ftp> quit
unix> zcat pdp.cns.94.1.ps.Z | lpr      # or however you print postscript

NOTE:  

The compressed file is 306994 bytes long.
Uncompressed, the file is 840184 byes long.

The printed version is 63 total pages long.

For those who do not have FTP access, physical copies can be requested from
Barbara Dorney <bd1q+ at andrew.cmu.edu>.  Ask for the report by title or
pdp.cns technical report number.






From bogner at eleceng.adelaide.edu.au  Tue Dec 13 23:22:37 1994
From: bogner at eleceng.adelaide.edu.au (Robert E. Bogner)
Date: Wed, 14 Dec 1994 14:52:37 +1030
Subject: [dwang@cis.ohio-state.edu: About sequential learning (or interference)]
Message-ID: <9412140422.AA16796@augean.eleceng.adelaide.edu.au>


From dwang at cis.ohio-state.edu  Mon Dec 12 10:50:23 1994
From: dwang at cis.ohio-state.edu (DeLiang Wang)
Date: Mon, 12 Dec 1994 10:50:23 -0500
Subject: About sequential learning (or interference)
Message-ID: <mailman.717.1149540302.24850.connectionists@cs.cmu.edu>

1. Visakan Kadirkamanathan wrote a dissertation on "Sequential
Learning in Artifician neural networks for PhD at U. of Cambridge,
October 1991. there were some conference papers undeer the same author.

2. C. J. Wallace, "Autonomous Rehearsal: Stability and plasticity in a
multilayer perceptron", project report, Dept. of EEE, Univ. of
Adelaide, October 1992  " . . a psychologically-motivated idea for
mainaining the original learning of a classifier neural network by
getting it to reinforce associations formed from confidently
classified random inputs. . ."

Robert E. Bogner

--------------------------------------------------------------------------
e_mail         bogner at eleceng.adelaide.edu.au             WATCH THIS
                                                            SPACE
Reliable mail: Robert E. Bogner
               Professor of Electrical Engineering,       
	       The University of Adelaide,          	  
	       Adelaide 5005,                       	  
	       SOUTH AUSTRALIA                      	  

Phone 	+61 8 303 5589 (answering machine all hours)    
Fax  	+61 8 303 4360  OR +61 8 224 0464
Home phone +61 8 332 5549
Time GMT+9h30m in northern hemisphere summer
     GMT+10h30m in northern hemisphere winter
==========================================================================


From sef+ at cs.cmu.edu  Wed Dec 14 00:03:50 1994
From: sef+ at cs.cmu.edu (Scott E. Fahlman)
Date: Wed, 14 Dec 94 00:03:50 EST
Subject: About sequential learning (or interference) 
In-Reply-To: Your message of Tue, 13 Dec 94 21:35:22 -0500.
             <199412140235.VAA17037@lennon.cc.gatech.edu> 
Message-ID: <mailman.718.1149540302.24850.connectionists@cs.cmu.edu>


Yet another approach to incremental learning can be seen in the Cascade
Correlation algorithm.  This creates hidden units (or feature detectors, if
you prefer) one by one, and freezes each one after it is created.  The
weights between these features and the outputs remain plastic and continue
to be trained.

This means that if you train a net on training set A and then switch to a
different training set B, you may build some new hidden units for task B,
but the hidden units created for task A are not cannibalized.  The A units
may be used in task B, either directly or as inputs to new hidden units,
but they remain unchanged and available.  As task B is trained, the output
weights change, so performance on task A will generally decline, but it can
come back very quickly if task A is re-trained or if you now train with a
combined A+B task.  The point is that the time-consuming part of learning
is in creating the hidden units, and these are retained once they are
learned.

My Recurrent Cascade-Correlation paper has an example of this.  This
recurrent net can be trained to recognize a temporal sequence of 1's and
0's as Morse code.  If you train all 26 letter-codes as a single training
set, the network will learn the task, but learning is faster and the
resulting net is smaller if you break the training set up into "lessons" of
increasing difficulty: first train on the shortest codes, then the
medium-length ones, then the longest ones, and finally on the whole set
together.  This is reminiscent of the modular networks explored by Waibel
and his colleagues for large speech tasks, but in this case you just have
to chop the training set up into modules -- the network architecture takes
care of itself.

References (these are also available online on Neuroprose, among other
places):

Scott E. Fahlman and Christian Lebiere (1990) "The Cascade-Correlation
Learning Architecture", in {\it Advances in Neural Information Processing
Systems 2}, D. S. Touretzky (ed.), Morgan Kaufmann Publishers, Los Altos
CA, pp. 524-532.

Scott E. Fahlman (1991) "The Recurrent Cascade-Correlation Architecture" in
{\it Advances in Neural Information Processing Systems 3}, R. P. Lippmann,
J. E. Moody, and D. S. Touretzky (eds.), Morgan Kaufmann Pulishers, Los
Altos CA, pp. 190-196.

-- Scott

===========================================================================
Scott E. Fahlman			Internet:  sef+ at cs.cmu.edu
Principal Research Scientist		Phone:     412 268-2575
School of Computer Science              Fax:       412 268-5576 (new!)
Carnegie Mellon University		Latitude:  40:26:46 N
5000 Forbes Avenue			Longitude: 79:56:55 W
Pittsburgh, PA 15213			Mood:      :-)
===========================================================================


From pollack at cs.brandeis.edu  Wed Dec 14 14:45:33 1994
From: pollack at cs.brandeis.edu (Jordan Pollack)
Date: Wed, 14 Dec 1994 14:45:33 -0500
Subject: Apology for Flame, and Survey 
In-Reply-To: <Pine.SGI.3.90.941213195922.11172B-100000@umbc8.umbc.edu> (bisant@gl.umbc.edu)
Message-ID: <199412141945.OAA10961@garnet.cs.brandeis.edu>

I flamed about some of the policies of WCNN, and David Bisant wrote:

> new problems
>in conferencing appear.  I think their fee structure is a reasonable 
>approach to solving these problems.

I obviously over-reacted, since I agree that there is also a more
positive reading of these policies which I didnt consider at the time -
that they are innovations that that the organization is using to cope
with several problems, such as:

a) filtering out computer-generated articles,
b) authors not showing up to deliver talks or posters,
c) expense of supporting travel costs for both plenary 
   speakers and conference organizers, and
d) declining membership/subscription revenues.

It was the simultaneous appearance of all these novel structures in
one place which raised a red flag, and led me to breathe fire.  I am
not usually so rude, and now I owe MANY colleagues apologies for my
outburst.

First, I'm very sorry for linking the two completely separate
societies - IEEE and INNS, and for denigrating the papers of
individuals who have published their work in these refereed meetings
over the years. Also, I want to apologize to the dozens of major
scientists involved in organizing and governing INNS, for sullying
their good names, and finally to Mike Cohen and my other new neighbors
at Boston Univ, who are merely using their good offices to promote the
meeting.

I hope the society will eventually forgive my unsociable outburst.

However, I do believe the $35 reviewing fee is a serious and big
issue, as I have never seen a precedent in any branch of science. And
while there are real costs to reviewing which are bundled into
the costs of meetings, this "unbundling" is a very slippery slope
to walk. If such fees catch on in conferences, they might catch on
universally, from journals to funding agencies to faculty search
committees, and this would dramatically change the "social contract,"
I fear, with many more negative than positive results!  But perhaps
even in this sub-issue I am over-reacting. Since the list is too large for
a discussion, I would like to perform an informal survey, which I
will then condense and post the results:

1) What do you think about review fees for scientific conferences?

2) What about review fees for scientific journals?

3) What about review fees for grant applications to NSF/NIH and other agencies?

4) What about review fees for faculty and postdoc job applications?

5) What do you predict as the ramifications of changing the current
scientific free review system to a fee-based review system?

Jordan Pollack                           Associate Professor
Computer Science Department              Center for Complex Systems
Brandeis University                      Phone: (617) 736-2713/fax 2741
Waltham, MA 02254                        email: pollack at cs.brandeis.edu




From white at pwa.acusd.edu  Wed Dec 14 14:44:50 1994
From: white at pwa.acusd.edu (Ray White)
Date: Wed, 14 Dec 1994 11:44:50 -0800 (PST)
Subject: FLAME: WCNN'95 - science in the mud pits
In-Reply-To: <199412081651.LAA01963@garnet.cs.brandeis.edu>
Message-ID: <Pine.3.89.9412141127.A16497-0100000@pwa.acusd.edu>

I'm not thrilled with the "reading fee" either, but $35 plus early
member registration @ $170 is pretty low for a conference these days;
certainly far less than the maximum registration fee of $480 that
Jordan Pollack mentioned.

As far as "sorting papers by fame" goes, for those of us who are not
famous, it is a good thing that there are conferences where papers
are merely "sorted by fame", when there are also elitist conferences
where fame is more likely a requirement for acceptance. (Did anyone
read NIPS into that sentence?) A conference is remembered for the quality
of presented papers, not for the quality of rejected papers.

Ray H. White			|  Depts. of Physics and Computer Science
<white at teetot.acusd.edu>	|  University of San Diego
619/260-4244 or 619/260-4627	|  5998 Alcala Park,  San Diego, CA 92110







From jose at scr.siemens.com  Wed Dec 14 15:20:45 1994
From: jose at scr.siemens.com (Stephen Hanson)
Date: Wed, 14 Dec 1994 15:20:45 -0500 (EST)
Subject: About sequential learning (or interference)
In-Reply-To: <199412140235.VAA17037@lennon.cc.gatech.edu>
References: <199412140235.VAA17037@lennon.cc.gatech.edu>
Message-ID: <YivpCRu1GEMnA0vvJt@tractatus.siemens.com>


of course avoiding "interference" is
another way of preventing generalization.

As usual there is a tradeoff here.

Steve




Stephen J. Hanson, Ph.D.
Head, Learning Systems Department
SIEMENS Research 
755 College Rd. East
Princeton, NJ 08540



From lautrup at connect.nbi.dk  Thu Dec 15 11:56:25 1994
From: lautrup at connect.nbi.dk (Benny Lautrup)
Date: Thu, 15 Dec 94 11:56:25 MET
Subject: New preprint: "Massive Weight Sharing: A Cure for ..."
Message-ID: <mailman.719.1149540302.24850.connectionists@cs.cmu.edu>

Subject: Paper available: "Massive Weight Sharing: A Cure for ..."
Date: August 29, 1994

FTP-host: connect.nbi.dk
FTP-file: neuroprose/lautrup.massive.ps.Z

----------------------------------------------

The following paper is now available:

Massive Weight Sharing: A Cure for Extremely Ill-posed Learning [12 pages]

B. Lautrup, L.K. Hansen, I. Law, N. Moerch, C. Svarer and S.C. Strother
CONNECT, The Niels Bohr Institute, University of Copenhagen, Denmark

Presented at the workshop on Supercomputing in Brain Research: From Tomography
to Neural Networks, HLRZ, Juelich, Germany, November 21-23, 1994

Abstract:
In most learning problems, adaptation to given examples is  well-posed  because
the number of examples far exceeds the number of  internal  parameters  in  the
learning machine. Extremely ill-posed learning problems are, however, common in
image and spectral analysis. They are characterized by a vast number of  highly
correlated inputs, e.g. pixel or pin values, and a modest number  of  patterns,
e.g. images or spectra. In this paper we show, for the case of  a  set  of  PET
images differing only in the values of  one  stimulus  parameter,  that  it  is
possible to train a neural network to learn the underlying rule  without  using
an excessive number of network weights or large amounts of computer  time.  The
method is based upon the observation that the standard learning rules  conserve
the subspace spanned by the input images.

Please do not reply directly to this message.

-----------------------------------------------

FTP-instructions:

unix> ftp connect.nbi.dk (or 130.225.212.30)
ftp> Name: anonymous
ftp> Password: your e-mail address
ftp> cd neuroprose
ftp> binary
ftp> get lautrup.massive.ps.Z
ftp> quit
unix> uncompress lautrup.massive.ps.Z
                                             
-----------------------------------------------

Benny Lautrup,
Computational Neural Network Center (CONNECT)
Niels Bohr Institute
Blegdamsvej 17
2100 Copenhagen
Denmark

Telephone: +45-3532-5200
Direct:    +45-3532-5358
Fax:       +45-3142-1016
e-mail:    lautrup at connect.nbi.dk


From pah at unixg.ubc.ca  Thu Dec 15 14:03:03 1994
From: pah at unixg.ubc.ca (Phil A. Hetherington)
Date: Thu, 15 Dec 1994 11:03:03 -0800 (PST)
Subject: sequential learning
In-Reply-To: <9412140416.AA17263@crab.psy.cmu.edu.psy.cmu.edu>
Message-ID: <Pine.SUN.3.91.941215105220.19696A-100000@unixg.ubc.ca>

A couple notes on interfernce. First, most of Jay's idea is still simply
theory, a good theory, but there is still little direct evidence that the
hippocampus does either pattern separation or pattern completion.  Most of
the pattern completion idea comes from Marr's conjectures on the
architecture of CA3 in hippocampus.  Second (to Hanson), interference is
not necessarily the flip side of generalization.  This is intuitive, but you
can reduce one without affecting the other.  See Lewandowsky's stuff on
this .  Next, whether the interference observed in connectionist nets is a
problem or not depends on your needs.  Is the net to model human learning
or to solve a particular engineering problem.  If the former, then see
McRae and Hetherington (1993)--it is not clear that realisticly designed
nets suffer from catastrophic interference.  Finally, it appears the the
bigger problem is to reduce interference without eliminating the
distinction between old and new items.  This was Ratcliff's main concern,
and is addressed by Sharkey. 

See also:

French, R.M. (1991). Proceedings of the Thirteenth Annual Conference of
  the Cognitive Science Society, 173-178. 
French, R.M. (1992). Connection Science, 4, 3-4, 365-377. 
Hetherington, P.A. (1990).  Neural Network Review, 4(1), 27-29. 
Hetherington, P.A. (1991).  The Sequential Learning Problem in 
  Connectionist Networks.  Unpublished Masters Thesis, McGill
  University, Montreal, Quebec. 
Hetherington, P.A., & Seidenberg, M.S. (1989).  Proceedings of the 
  Eleventh Annual Conference of the Cognitive Science Society, 26-33. 
Hinton, G.E., & Plaut, D.C. (1987).  Proceedings of the Ninth Annual 
  Conference of the Cognitive Science Society, 177-186.
Kortge, C.A. (1990). Proceedings of the Twelfth Annual Conference of the
  Cognitive Science Society, 764-771. 
Krushke, J.K. (1992).  Psychological Review, 99, 22-44. 
Krushke, J.K. (1993).  Connection Science, 5, 3-36.
Lewandowsky, S. (1991).  In W.E. Hockley & S, Lewandowsky (Eds.), Relating
  theory and data: Essays on human memory in honor of Bennet B. Murdock.
McRae, K., & Hetherington, P.A. (1993). Proceedings of the Fifteenth
  Annual Conference of the Cognitive Science Society, 723-728. 
Murre, J.M.J. (1992).  Learning and categorization in modular neural 
  networks.  Lawrence Erlbaum. 
Sharkey, N.E., & Sharkey, A.J.C. Understanding Catastrophic
  Interference in neural Nets.  Technical Report, Department of Computer
  Science, University of Sheffield, U.K. 
Sloman, S.A., & Rumelhart, D.E. (1992).  In A. Healy, S.M. Kosslyn, & 
  R.M. Shiffrin (Eds.), From learning theory to cognitive processes: Essays 
  in honor of William K. Estes. 





From french at cogsci.indiana.edu  Thu Dec 15 14:28:01 1994
From: french at cogsci.indiana.edu (Bob French)
Date: Thu, 15 Dec 94 14:28:01 EST
Subject: Catastrophic forgetting and sequential learning
Message-ID: <mailman.720.1149540302.24850.connectionists@cs.cmu.edu>


Below I have indicated a number of references on current work on the
problem of sequential learning in connectionist networks.  All of these
papers address the problem of catastrophic interference, which may
result when a previously trained connectionist network attempts to
learn new patterns.
   The following commented list is by no means complete.  In
particular, no mention is made of convolution-correlation models with
their associated connectionist networks.  Nonetheless, I hope it
might prove to be a useful introduction to people interested in
knowing a bit more about the subject.

				Bob French
                                french at cogsci.indiana.edu


----------------------------------------------------------------------------

                   Recent Work in Catastrophic Interference 
                          in Connectionist Networks


The two papers that really kicked off research in this area were:

McCloskey, M. & Cohen, N. (1989)  "Catastrophic interference in connections
	networks: the sequential learning problem" The Psychology of 
	Learning and Motivation, 24, 109-165.

Ratcliff, R (1990) "Connectionist models of recognition memory: constraints
	imposed by learning and forgetting functions" Psychological Review,
	97, 285-308


Hetherington and Seidenberg very early suggested an Ebbinghaus-like
"savings" measure of catastrophic interference and, based on this,
they concluded that catastrophic interference wasn't really as much of
a problem as had been thought. While the problem has, in fact, been
subsequently confirmed to be quite serious, the "savings" measure they
proposed is still widely used to measure the extent of forgetting.

Hetherington, P.. & Seidenberg, M. (1989) "Is there 'catastrophic
	interference' in connectionist networks?"  Proceedings of the
	11th Annual Conference of the Cognitive Science Society. 
	Hillsdale, NJ: Erlbaum, 26-33.


Kortge was one of the first to propose a solution to this problem,
using what he called "novelty vectors".

Kortge, C. (1990) "Episodic Memory in Connectionist Networks" Proceedings 
	of the 12th Annual Conference of the Cognitive Science Society. 
	Hillsdale, NJ: Erlbaum, 764-771


Sloman and Rumelhart also developed a technique called "episodic gating"
designed to reduce the severity of the problem.

Sloman, S. & Rumelhart, D., (1991) "Reducing interference in distributed
	memories through episodic gating" In Healy, Kosslyn, & Shiffrin (eds.)
	Essays in Honor of W. K. Estes.


In 1991 French suggested that catastrophic interference might be the 
inevitable price you pay for the advantages of fully distributed
representations (in particular, generalization).  He suggested a way
of dynamically producing "semi-distributed" hidden-layer
representations to reduce the effect of catastrophic interference.

French, R. (1991) "Using semi-distributed representations to overcome 
	catastrophic forgetting in connectionist networks" in Proceedings 
	of the 13th Annual Conference of the Cognitive Science Society. 
	Hillsdale, NJ: Erlbaum, 173-178.

A more detailed article presenting the same techique, called
activation sharpening, appeared in Connection Science.

French, R. (1992) "Semi-distributed Representations and Catastrophic
	Forgetting in Connectionist Networks", Connection Science,
	Vol. 4: 365-377.

In a more recent paper (1994), French presented a technique called
context biasing, which again dynamically "massages" hidden layer
representations based on the "context" of other recently learned
exemplars.  The goal of this technique is to produce hidden-layer
representations that are simultaneously well distributed and
orthogonal.

French, R. (1994) "Dynamically constraining connectionist networks to
	produce distributed, orthogonal representations to reduce
	catastrophic interference" in Proceedings of the 16th Annual 
	Conference of the Cognitive Science Society. Hillsdale, NJ: 
	Erlbaum, 335-340

Finally, still in this vein, French proposed (1993) at the NIPS-93
workshop on catastrophic interference a dynamic system of two
interacting networks working in tandem, one storing prototypes, the
other doing short-term learning of new exemplars. For a "real brain"
justification for this type of architecture see McClelland,
McNaughton, and O'Reilly (1994), below.  (A full paper on this tandem
network architecture will be available at the beginning of next year.)
This technique is discussed very briefly in:

French, R. (1994) "Catastrophic interference in connectionist networks:
	Can it be predicted, can it be prevented?" Neural Information
	Processing Systems - 6, Cowan, Tesauro, Alspector (eds.) 
	San Francisco, CA: Morgan Kaufmann. 1176-1177.


Steve Lewandowsky has also been very active in this area.  He
developed a simple technique in 1991 that focused on producing
orthogonalization at the input layer rather than the hidden layer.
This "symmetric vectors" technique is discussed in:

Lewandowsky, S. & Shu-Chen Li (1993) "Catastrophic Interference in
	Neural Networks: Causes, solutions and data" in New 
	Perspectives on interference and inhibition in cognition.
	Dempster & Brainerd (eds.) New York: NY: Academic Press.

Lewandowsky, S (1991) "Gradual unlearning and catastrophic
	interference: a comparison of distributed architectures.
 	In Hockley & Lewandowsky (eds.) Relating theory and
	data: Essays on human memory in honor of Bennet B.
	Murdock. (pp. 445-476).  Hillsdale, NJ: Lawrence Erlbaum.

and in an earlier University of Oklahoma psychology department
technical report:

Lewandowsky, S. (1993) "On the relation between catastrophic
	interference and generalization in connectionist networks"


In 1993 McRae and Hetherington published a study using pre-training to
eliminate catastrophic interference.

McRae, K. & Hetherington, P. (1993) "Catastrophic interference is 
	eliminated in pretrained networks" in Proceedings of the
	15th Annual Cognitive Science Society.  Hillsdale, NJ: 
	Erlbaum. 723-728


John Kruschke discussed the problem of catastrophic forgetting at
length in the context of his connectionist model, ALCOVE, and showed
the extent to which and under what circumstances this model is not
subject to catastrophic forgetting.

Kruschke, J. (1993) "Human category learning: implications for
	backpropagation models", Connection Science, Vol. 5, No. 1, 


Jacob Murre has also examined how his model, CALM, performs on
the sequential learning problem.    See, in particular:

Murre, J. Learning and Categorization in Modular neural networks
	Hillsdale, NJ: Lawrence Erlbaum. 1992. (see esp. ch. 7.4)


It is to be noted that both ALCOVE and CALM rely, at least in part,
on reducing the distributedness of their internal representations in
order to achieve improved performance on the problem of catastrophic
interference.


A 1994 article (in press) by Anthony Robins presents a novel
technique, called "pseudorehearsal", whereby "pseudoexemplars" that
reflect prior learning are added to the new data set to learned in
order to reduce catastrophic forgetting.

Robins, A. "Catastrophic forgetting, rehearsal, and pseudorehearsal",
	University of Otago (New Zealand) computer science technical
	report.  (copies:  coscavr at otago.ac.nz)


Tetewsky, Shultz & Buckingham (1994) demonstrate the improvements that
result from using Fahlman's cascade-correlation learning algorithm.

Tetewsky, S., Shultz, T. and Buckingham, D. "Assessing interference and
	savings in connectionist models of human recognition memory"
	Department of Psychology TR, McGill University, Montreal.
	(presented at 1994 Meeting of the Psychonomic Society).


Sharkey & Sharkey (1993) discussed the relation between the problem of
interference and discrimination in connectionist networks.  They
conclude that sequentially trained networks using backprop will
unavoidably suffer from one or the other problem.  I am not aware if
there is a final version of this paper in print yet, but Noel Sharkey
is currently at University of Sheffield, Dept. of Computer Science,
Sheffield.  n.sharkey at dcs.shef.ac.uk

Sharkey, N. & Sharkey, A., "An interference-discrimination tradeoff
	in connectionist models of human memory"  


McClelland, McNaughton, & O'Reilly issued a CMU technical report
earlier this year (1994) in which they discuss the phenomenon of
catastrophic interference in the context of the "real world", i.e.,
the brain.  They suggest that the complementary learning systems in
the hippocampus and the neocortex might be the brain's way of 
overcoming the problem.  They argue that this dual system provides 
a means not only of rapidly acquiring new information, but also of 
storing well-learned information as prototypes.

McClelland, J., McNaughton, B., & O'Reilly, R. "Why there are
	complementary learning systems in the hippocampus and 
	neocortex:  Insights from the successes and failures
	of connectionist models of learning and memory"  CMU
	Tech report: PNP.CNS.94.1, March 1994.

-------------------------------------------------------------------------




From marney at ai.mit.edu  Thu Dec 15 18:19:23 1994
From: marney at ai.mit.edu (Marney Smyth)
Date: Thu, 15 Dec 94 18:19:23 EST
Subject: NNSP95 : Call for Papers
Message-ID: <9412152319.AA00894@motor-cortex>

**********************************************************************
*                  1995 IEEE WORKSHOP ON			     *
*   								     *
*           NEURAL NETWORKS FOR SIGNAL PROCESSING		     *
*								     *
*     August 31 -- September 2, 1995, Cambridge, Massachusetts, USA  *
*          Sponsored by the IEEE Signal Processing Society           *
*       (In cooperation with the IEEE Neural Networks Council)       *
*								     *
**********************************************************************


                  FIRST ANNOUNCEMENT AND CALL FOR PAPERS
 
Thanks to the sponsorship of IEEE Signal Processing Society, the
co-sponsorship of IEEE Neural Network Council, the fifth of a series
of IEEE Workshops on Neural Networks for Signal Processing will be
held at the Royal Sonesta Hotel, Cambridge, Massachusetts, on Thursday
8/31 -- Saturday 9/2, 1995. Papers are solicited for, but not limited
to, the following topics:

++ APPLICATIONS:
Image, speech, communications, sensors, medical, adaptive
filtering, OCR, and other general signal processing and pattern
recognition topics.

++ THEORIES:
Generalization and regularization, system identification, parameter
estimation, new network architectures, new learning algorithms, and
wavelets in NNs. 

++ IMPLEMENTATIONS:
Software, digital, analog, hybrid technologies, parallel processing.


Prospective authors are invited to submit 5 copies of extended summaries
of no more than 6 pages. The top of the first page of the summary should
include a title, authors' names, affiliations, address, telephone and
fax numbers and e-mail address if any. Camera-ready full papers
of accepted proposals will be published in a hard-bound volume by IEEE
and distributed at the workshop.  For further information, please
contact Marney Smyth (Tel.) (617)-253-0547, (Fax) (617)-253-2964
(e-mail) marney at ai.mit.edu. 

We plan to use the World Wide Web (WWW) for posting further announcements 
on NNSP95 such as: submitted papers status, final program, hotel information 
etc. You can use MOSAIC and access URL site: http://www.cdsp.neu.edu.
If you do not have access to WWW use anonymous ftp to site ftp.cdsp.neu.edu
and look under the directory /pub/NNSP95.

Please send paper submissions to:

Prof. Elias S. Manolakos 
IEEE  NNSP'95
409 Dana Research Building 
Electrical and Computer Engineering Department 
Northeastern University, Boston, MA 02115, USA 

Phone: (617) 373-3021,  Fax: (617) 373-4189                  

*******************
* IMPORTANT DATES *
*******************

Extended summary received by:            February 17

Notification of acceptance:              April 21

Photo-ready accepted papers received by: May 22 

Advanced registration received before:   June 2 

                                    
GENERAL CHAIRS

Federico Girosi
Center for Biological and Computational Learning, MIT
girosi at ai.mit.edu

John Makhoul
Bolt, Beranek and Newman
makhoul at bbn.com


PROGRAM CHAIR

Elias S. Manolakos
Communications and Digital Signal Processing (CDSP)
Northeastern University
elias at cdsp.neu.edu

FINANCE CHAIR			LOCAL ARRANGEMENTS

Judy Franklin                   Mary Pat Fitzgerald, MIT
GTE Laboratories		marypat at ai.mit.edu
jfranklin at gte.com

PUBLICITY CHAIR			PROCEEDINGS CHAIR

Marney Smyth, MIT		Elizabeth J. Wilson
marney at ai.mit.edu		Raytheon Co. 			
				email: bwilson at sud2.ed.ray.com


TECHNICAL PROGRAM COMMITTEE

Joshua Alspector		John Makhoul
Alice Chiang			Elias Manolakos
A. Constantinides		P. Mathiopoulos 
Lee Giles			Nahesan Niranjan
Federico Girosi			Tomaso Poggio
Lars Kai Hansen 		Jose Principe
Yu-Hen Hu			Wojtek Przytula
Jenq-Neng Hwang			John Sorensen
Bing-Huang Juang		Andreas Stafylopatis
Shigeru Katagiri		John Vlontzos
George Kechriotis		Raymond Watrous
Stephanos Kollias		Christian Wellekens
Sun-Yuan Kung			Ron Williams
Gary M. Kuhn			Barbara Yoon
Richard Lippmann		Xinhua Zhuang 





From Dave_Touretzky at DST.BOLTZ.CS.CMU.EDU  Fri Dec 16 03:22:31 1994
From: Dave_Touretzky at DST.BOLTZ.CS.CMU.EDU (Dave_Touretzky@DST.BOLTZ.CS.CMU.EDU)
Date: Fri, 16 Dec 94 03:22:31 EST
Subject: elitism at NIPS
Message-ID: <1394.787566151@DST.BOLTZ.CS.CMU.EDU>

The folks involved in running NIPS are aware that in the past, some people
have felt the conference was biased against outsiders.  As the program
chair for NIPS*94, I want to mention some steps we took to reduce that
perception:

* This year we required authors to submit full papers, not just abstracts.
  That made it harder for "famous" people to get by on their reputation
  alone, and easier for "non-famous" people to get good papers in.

* This year we required reviewers to provide real feedback to all authors.
  In order to make this less painful, we completely redesigned the review
  form (it's publicy accessible via the NIPS home page) and, for the first
  time, accepted reviews by email.  Everyone liked getting comments from
  the reviewers, and authors whose papers were not accepted understood why.

* We continue to recruit new people for positions in both the organizing
  and program committees.  It's not the same dozen people year after year.
  We also have a large reviewer pool:  176 people served as reviewers
  this year.

* We tend to bring in "outsiders" as our invited speakers, rather than
  the usual good old boys.  This year's invited speakers included Francis
  Crick of the Salk Insititute, Bill Newsome (a neuroscientist from 
  Stanford), and Malcolm Slaney (a signal processing expert formerly with
  Apple and now at Interval Research.)  None had been to NIPS before.

The fundamental limits on NIPS paper acceptances are that (1) we're
committed to a single-track conference, and (2) we only have room for 138
papers in the proceedings.  Therefore the reviewing process has to be
selective.

The acceptance rate this year was 33%; it was 25% in past years.  The
change is due to fewer submissions, not more acceptances.  Requiring full
papers was probably the cause of the drop in submissions.  The quality of
the accepted papers has remained high.

There are other ways to participate in NIPS besides writing a paper.  We
have a very successful workshop program, where lots of intense interaction
takes place between people with similar interests.  Many people give talks
at the workshops and not at the conference.  (I did that this year.)  NIPS
issues a call for workshop proposals at about the same time as the call for
papers.  Consider organizing a workshop at NIPS*95, or at least
participating in one.  You do not even have to register for the conference;
workshop registration is entirely separate.

The URL for the NIPS home page appears below.  Besides the review form,
you'll also find formatting instructions for paper submissions.  Authors of
accepted NIPS*94 papers will find instructions for how to submit their
final camera-ready copy, which is due January 23.

  http://www.cs.cmu.edu:8001/afs/cs/project/cnbc/nips/NIPS.html

-- Dave Touretzky
   NIPS*94 Program Chair


From P.McKevitt at dcs.shef.ac.uk  Fri Dec 16 05:55:02 1994
From: P.McKevitt at dcs.shef.ac.uk (Paul Mc Kevitt)
Date: Fri, 16 Dec 94 10:55:02 GMT
Subject: AISB-95 workshop call for papers
Message-ID: <mailman.721.1149540302.24850.connectionists@cs.cmu.edu>


<<FIRST CALL FOR PAPERS>>  <<FIRST CALL FOR PAPERS>>  <<FIRST CALL FOR PAPERS>>
<<FIRST CALL FOR PAPERS>>  <<FIRST CALL FOR PAPERS>>  <<FIRST CALL FOR PAPERS>>

                            Advance Announcement
  
                  FIRST CALL FOR PAPERS AND PARTICIPATION

                            AISB-95 Workshop on

                             REACHING FOR MIND:
                      FOUNDATIONS OF COGNITIVE SCIENCE

                            April 3rd/4th 1995

                                  at the
       The Tenth Biennial Conference on AI and Cognitive Science (AISB-95)
                   (Theme: Hybrid Problems, Hybrid Solutions)

                                 Halifax Hall
                           University of Sheffield
                              Sheffield, England
                    (Monday 3rd -- Friday 7th April 1995)
 
                          Society for the Study of
        Artificial Intelligence and Simulation of Behaviour (SSAISB)

                                   Chair:
                              Sean O Nuallain                              

                 Dublin City University, Dublin, Ireland &
                 National Research Council, Ottawa, Canada

                                 Co-Chair:
                               Paul Mc Kevitt

                       Department of Computer Science
                      University of Sheffield, England



WORKSHOP COMMITTEE:

John Barnden      (New Mexico State University, NM, USA)
Istvan Berkeley   (University of Alberta, Canada)
Mike Brady        (Oxford, England)
Harry Bunt        (ITK, Tilburg, The Netherlands)
Peter Carruthers  (University of Sheffield, England) 
Daniel Dennett    (Tufts University, USA)
Eric Dietrich     (SUNY Binghamton, NY, USA)
Jerry Feldman     (ICSI, UC Berkeley, USA)
John Frisby       (University of Sheffield, England)
Stevan Harnad     (University of Southampton, England)
James Martin      (University of Colorado at Boulder, CO, USA)
John Macnamara    (McGill University, Canada)
Mike McTear       (Universities of Ulster and Koblenz, Germany)
Ryuichi Oka       (RWC P, Tsukuba, Japan)
Jordan Pollack    (Ohio State University, OH, USA)
Zenon Pylyshyn    (Rutgers University, USA)
Ronan Reilly      (University College, Dublin, Ireland)
Roger Schank      (ILS, Illinois)
NNoel Sharkey      (University of Sheffield, England)
Walther v.Hahn    (University of Hamburg, Germany)
Yorick Wilks      (University of Sheffield, England)


WORKSHOP DESCRIPTION

The assumption underlying this workshop is that Cognitive Science (CS)
is in crisis.  The crisis manifests itself, as exemplified by the
recent Buffalo summer institute, in a complete lack of consensus among
even the biggest names in the field on whether CS has or indeed should
have a clearly identifiable focus of study; the issue of identifying
this focus is a separate and more difficult one. Though academic
programs in CS have in general settled into a pattern compatible with
classical computationalist CS (Pylyshyn 1984, Von Eckardt 1993),
including the relegation from focal consideration of consciousness,
affect and social factors, two fronts have been opened on this
classical position.

The first front is well-publicised and highly visible. Both Searle
(1992) and Edelman (1992) refuse to grant any special status to
information-processing in explanation of mental process. In contrast,
they argue, we should focus on Neuroscience on the one hand and
Consciousness on the other. The other front is ultimately the more
compelling one. It consists of those researchers from inside CS who
are currently working on consciousness, affect and social factors and
do not see any incompatibility between this research and their vision
of CS, which is that of a Science of Mind (see Dennett 1993, O
Nuallain (in press) and Mc Kevitt and Partridge 1991, Mc Kevitt and Guo
1994).

References

Dennett, D. (1993) Review of John Searle's "The Rediscovery of the
Mind". The Journal of Philosophy 1993, pp 193-205

Edelman, G.(1992) Bright Air, Brilliant Fire. Basic Books

Mc Kevitt, P. and D. Partridge (1991) Problem description and
hypothesis testing in Artificial Intelligence In ``Artificial
Intelligence and Cognitive Science '90'', Springer-Verlag British
Computer Society Workshop Series, McTear, Michael and Norman Creaney
(Eds.), 26-47, Berlin, Heidelberg: Springer-Verlag.  Also, in
Proceedings of the Third Irish Conference on Artificial Intelligence
and Cognitive Science (AI/CS-90), University of Ulster at Jordanstown,
Northern Ireland, EU, September and as Technical Report 224,
Department of Computer Science, University of Exeter, GB- EX4 4PT,
Exeter, England, EU, September, 1991.

Mc Kevitt, P. and Guo, Cheng-ming (1995) From Chinese rooms to Irish
rooms: new words on visions for language. Artificial Intelligence
Review Vol. 8. Dordrecht, The Netherlands: Kluwer-Academic Publishers.
(unabridged version) First published: International Workshop on
Directions of Lexical Research, August, 1994, Beijing, China.

O Nuallain, S (in press) The Search for Mind: a new foundation for
CS. Norwood: Ablex

Pylyshyn, Z.(1984) Computation and Cognition. MIT Press

Searle, J (1992) The rediscovery of the mind. MIT Press.

Von Eckardt, B. (1993) What is Cognitive Science? MIT Press

WORKSHOP TOPICS:

The tension which riddles current CS can therefore be stated thus: CS,
which gained its initial capital by adopting the computational
metaphor, is being constrained by this metaphor as it attempts to
become an encompassing Science of Mind. Papers are invited for this
workshop which:


* Address the central tension

* Propose an overall framework for CS (as attempted, inter alia,
  by O Nuallain (in press))

* Explicate the relations between the disciplines which comprise CS.

* Relate educational experiences in the field

* Describe research outside the framework of classical
  computationalist CS in the context of an alternative framework

* Promote a single logico-mathematical formalism as a theory of
  Mind (as attempted by Harmony theory)

* Disagree with the premise of the workshop


Other relevant topics include:

* Classical vs. neuroscience representations

* Consciousness vs. Non-consciousness

* Dictated vs. emergent behaviour

* A life/Computational intelligence/Genetic algorithms/Connectionism

* Holism and the move towards Zen integration


The workshop will focus on three themes:

* What is the domain of Cognitive Science ?

* Classic computationalism and its limitations

* Neuroscience and Consciousness


WORKSHOP FORMAT:

Our intention is to have as much discussion as possible during the
workshop and to stress panel sessions and discussion rather than
having formal paper presentations.  The workshop will consist of
half-hour presentations, with 15 minutes for discussion at the end of
each presentation and other discussion sessions.  A plenary session at
the end will attempt to resolve the themes emerging from the different
sessions.

ATTENDANCE:

We hope to have an attendance between 25-50 people at the workshop.

Given the urgency of the topic, we expect it to be of interest not
only to scientists in the AI/Cognitive Science (CS) area, but also to
those in other of the sciences of mind who are curious about CS.
We envisage researchers from Edinburgh, Leeds, York, Sheffield and Sussex
attending from within England and many overseas visitors as the
Conference Programme is looking very international.


SUBMISSION REQUIREMENTS:

Papers of not more than 8 pages should be submitted by electronic mail
(preferably uuencoded compressed postscript) to Sean O Nuallain at the
E-mail address(es) given below.  If you cannot submit your paper by
E-mail please submit three copies by snail mail.

*******Submission Deadline:  February 13th 1995
*******Notification Date:    February 25th 1995
*******Camera ready Copy:    March 10th 1995


PUBLICATION:

Workshop notes/preprints will be published.  If there is sufficient
interest we will publish a book on the workshop possibly with the
American Artificial Intelligence Association (AAAI) Press.

WORKSHOP CHAIR:

Sean O Nuallain  
((Before Dec 23:))
Knowledge Systems Lab,                       
Institute for Information Technology,
National Research Council,
Montreal Road, Ottawa
Canada K1A OR6

Phone:  1-613-990-0113
E-mail: sean at ai.iit.nrc.ca
FaX:    1-613-95271521

((After Dec 23:))
Dublin City University,
IRL- Dublin 9, Dublin
Ireland, EU

WWW:   http://www.compapp.dcu.ie
Ftp:   ftp.vax1.dcu.ie
E-mail: onuallains at dcu.ie
FaX: 353-1-7045442
Phone: 353-1-7045237


AISB-95 WORKSHOPS AND TUTORIALS CHAIR:

    Dr. Robert Gaizauskas
    Department of Computer Science
    University of Sheffield
    211 Portobello Street
    Regent Court
    Sheffield S1 4DP
    U.K.  E-mail:  robertg at dcs.shef.ac.uk
    WWW:     http://www.dcs.shef.ac.uk/
    WWW:     http://www.shef.ac.uk/
    Ftp:     ftp.dcs.shef.ac.uk
    FaX:     +44 (0) 114 278-0972
    Phone:   +44 (0) 114 278-5572


AISB-95 CONFERENCE/LOCAL ORGANISATION CHAIR:

Paul Mc Kevitt
Department of Computer Science
Regent Court
211 Portobello Street
University of Sheffield
GB- S1 4DP, Sheffield
England, UK, EU.

E-mail:    p.mckevitt at dcs.shef.ac.uk
WWW:       http://www.dcs.shef.ac.uk/
WWW:       http://www.shef.ac.uk/
Ftp:       ftp.dcs.shef.ac.uk
FaX:       +44 (0) 114-278-0972
Phone:     +44 (0) 114-282-5572 (Office)   
                       282-5596 (Lab.)    
          	       282-5590 (Secretary)


AISB-95 REGISTRATION:

        Alison White
        AISB Executive Office
        Cognitive and Computing Sciences (COGS)
        University of Sussex
        Falmer, Brighton
        England, UK, BN1 9QH

        Email:	alisonw at cogs.susx.ac.uk
        WWW: http://www.cogs.susx.ac.uk/users/christ/aisb
        Ftp: ftp.cogs.susx.ac.uk/pub/aisb
        Tel:	+44 (0) 1273 678448
        Fax:	+44 (0) 1273 671320


AISB-95 ENQUIRIES:

        Debbie Daly,
        Administrative Assistant, AISB-95,
	Department of Computer Science,
        Regent Court,
        211 Portobello Street,
	University of Sheffield,
	GB- S1 4DP, Sheffield,
        UK, EU.
	
	Email:    debbie at dcs.shef.ac.uk (personal communication)
	Fax:      +44 (0) 114-278-0972
	Phone:    +44 (0) 114-278-5565                (personal)
                                 -5590                (messages)

	Email:    aisb95 at dcs.shef.ac.uk                    (for auto responses)
        WWW: http://www.dcs.shef.ac.uk/aisb95      [Sheffield Computer Science]
        Ftp: ftp.dcs.shef.ac.uk (cd aisb95)
        WWW: http://www.shef.ac.uk/              [Sheffield Computing Services]
        Ftp: ftp.shef.ac.uk (cd aisb95)
        WWW: http://ijcai.org/)  (Email welty at ijcai.org)   [IJCAI-95, MONTREAL]
        WWW: http://www.cogs.susx.ac.uk/users/christ/aisb [AISB SOCIETY SUSSEX]
        Ftp: ftp.cogs.susx.ac.uk/pub/aisb 


VENUE:

The venue for registration and all conference events is: 

	Halifax Hall of Residence,
	Endcliffe Vale Road,
	GB- S10 5DF, Sheffield,
        UK, EU.

	FaX: +44 (0) 114-268-4227	
	Tel: +44 (0) 114-268-2758 (24 hour porter)
	Tel: +44 (0) 114-266-4196 (manager)


SHEFFIELD:

Sheffield is one of the friendliest cities in Britain and is situated
well having the best and closest surrounding countryside of any major
city in the UK. The Peak District National Park is only minutes
away. It is a good city for walkers, runners, and climbers.  It has
two theatres, the Crucible and Lyceum. The Lyceum, a beautiful
Victorian theatre, has recently been renovated.  Also, the city has
three 10 screen cinemas.  There is a library theatre which shows more
artistic films. The city has a large number of museums many of which
demonstrate Sheffield's industrial past, and there are a number of
Galleries in the City, including the Mapping Gallery and Ruskin.  A
number of important ancient houses are close to Sheffield such as
Chatsworth House. The Peak District National Park is a beautiful site
for visiting and rambling upon.  There are large shopping areas in the
City and by 1995 Sheffield will be served by a 'supertram' system: the
line to the Meadowhall shopping and leisure complex is already open.

The University of Sheffield's Halls of Residence are situated on the
western side of the city in a leafy residential area described by John
Betjeman as ``the prettiest suburb in England''.  Halifax Hall is
centred on a local Steel Baron's house, dating back to 1830 and set in
extensive grounds. It was acquired by the University in 1830 and
converted into a Hall of Residence for women with the addition of a
new wing.


ARTIFICIAL INTELLIGENCE AT SHEFFIELD:

Sheffield Computer Science Department has a strong programme in
Cognitive Systems and is part of the University's Institute for
Language, Speech and Hearing (ILASH).  ILASH has its own machines and
support staff, and academic staff attached to it from nine
departments. Sheffield Psychology Department has the Artificial
Intelligence Vision Research Unit (AIVRU) which was founded in 1984 to
coordinate a large industry/university Alvey research consortium
working on the development of computer vision systems for autonomous
vehicles and robot workstations.  Sheffield Philosophy Department has
the Hang Seng Centre for Cognitive Studies, founded in 1992, which
runs a workshop/conference series on a two-year cycle on topics of
interdisciplinary interest. (1992-4: 'Theory of mind'; 1994- 6:
'Language and thought'.)






From lautrup at connect.nbi.dk  Fri Dec 16 12:01:20 1994
From: lautrup at connect.nbi.dk (Benny Lautrup)
Date: Fri, 16 Dec 94 12:01:20 MET
Subject: Paper available: "Massive Weight Sharing: A Cure for ..."
Message-ID: <mailman.722.1149540302.24850.connectionists@cs.cmu.edu>

FTP-host: connect.nbi.dk
FTP-file: neuroprose/lautrup.massive.ps.Z

----------------------------------------------

The following paper is now available:

Massive Weight Sharing: A Cure for Extremely Ill-posed Learning [12 pages]

B. Lautrup, L.K. Hansen, I. Law, N. Moerch, C. Svarer and S.C. Strother
CONNECT, The Niels Bohr Institute, University of Copenhagen, Denmark

Presented at the workshop on Supercomputing in Brain Research: From Tomography
to Neural Networks, HLRZ, Juelich, Germany, November 21-23, 1994

Abstract:
In most learning problems, adaptation to given examples is  well-posed  because
the number of examples far exceeds the number of  internal  parameters  in  the
learning machine. Extremely ill-posed learning problems are, however, common in
image and spectral analysis. They are characterized by a vast number of  highly
correlated inputs, e.g. pixel or pin values, and a modest number  of  patterns,
e.g. images or spectra. In this paper we show, for the case of  a  set  of  PET
images differing only in the values of  one  stimulus  parameter,  that  it  is
possible to train a neural network to learn the underlying rule  without  using
an excessive number of network weights or large amounts of computer  time.  The
method is based upon the observation that the standard learning rules  conserve
the subspace spanned by the input images.

Please do not reply directly to this message.

-----------------------------------------------

FTP-instructions:

unix> ftp connect.nbi.dk (or 130.225.212.30)
ftp> Name: anonymous
ftp> Password: your e-mail address
ftp> cd neuroprose
ftp> binary
ftp> get lautrup.massive.ps.Z
ftp> quit
unix> uncompress lautrup.massive.ps.Z

-----------------------------------------------

Benny Lautrup,
Computational Neural Network Center (CONNECT)
Niels Bohr Institute
Blegdamsvej 17
2100 Copenhagen
Denmark

Telephone: +45-3532-5200
Direct:    +45-3532-5358
Fax:       +45-3142-1016
e-mail:    lautrup at connect.nbi.dk


From jaap.murre at mrc-apu.cam.ac.uk  Fri Dec 16 08:46:00 1994
From: jaap.murre at mrc-apu.cam.ac.uk (Jaap Murre)
Date: Fri, 16 Dec 94 13:46:00 GMT
Subject: Papers and PC demo available
Message-ID: <9412161346.AA00917@rigel.mrc-apu.cam.ac.uk>


The following three files have recently (15-12-1994) been added to our
ftp site (ftp://ftp.mrc-apu.cam.ac.uk/pub/nn/murre):

File 1:
nnga1.ps   Happel, B.L.M., & J.M.J. Murre (1994). Design and
           evolution of modular neural network architectures. Neural
           Networks, 7, 985-1004. (About 0.5 Mb; ps.Z version is
           recommended.)

Abstract:  To investigate the relations between structure and function in
both artificial and natural neural networks, we present a series of
simulations and analyses with modular neural networks. We suggest a
number of design principles in the form of explicit ways in which neural
modules can cooperate in recognition tasks. These results may
supplement recent accounts of the relation between structure and
function in the brain. The networks used consist out of several modules,
standard subnetworks that serve as higher-order units with a distinct
structure and function. The simulations rely on a particular network
module called CALM (Murre, Phaf, and Wolters, 1989, 1992). This
module, developed mainly for unsupervised categorization and learning,
is able to adjust its local learning dynamics. The way in which modules
are interconnected is an important determinant of the learning and
categorization behaviour of the network as a whole. Based on arguments
derived from neuroscience, psychology, computational learning theory,
and hardware implementation, a framework for the design of such
modular networks is laid-out. A number of small-scale simulation
studies shows how intermodule connectivity patterns implement 'neural
assemblies' (Hebb, 1949) that induce a particular category structure in
the network. Learning and categorization improves as the induced
categories are more compatible with the structure of the task domain. In
addition to structural compatibility, two other principles of design are
proposed that underlie information processing in interactive activation
networks: replication and recurrence. 
      Because a general theory for relating network architectures to
specific neural functions does not exist, we extend the biological
metaphor of neural networks, by applying genetic algorithms (a
biocomputing method for search and optimization based on natural
selection and evolution) to search for optimal modular network
architectures for learning a visual categorization task. The best
performing network architectures seemed to have reproduced some of
the overall characteristics of the natural visual system, such as the
organization of coarse and fine processing of stimuli in separate
pathways. A potentially important result is that a genetically defined
initial architecture cannot only enhance learning and recognition
performance, but it can also induce a system to better generalize its
learned behaviour to instances never encountered before. This may
explain why for many vital learning tasks in organisms only a minimal
exposure to relevant stimuli is necessary.

File 2:
chaos1.ps  Happel, B.L.M., & J.M.J. Murre (submitted). Evolving
           complex dynamics in modular interactive neural networks.
           Submitted to Neural Networks. (This is a large file: 1.5 Mb!
           Retrieve the ps.Z version if possible.)

Abstract:  Computational simulation studies, carried out within a
general framework of modular neural network design, demonstrate that
networks consisting of many interacting modules provide a variety of
different neural processing principles. The dynamics underlying these
behaviors range from simple linear separation of input vectors in
individual modules, to oscillations, evoking chaotic regimes in the
activity evolution of a network. As opposed to static representations in
conventional neural network models, information in oscillatory networks
is represented as space-time patterns of activity. Chaos in a neural
network can serve as: (i) a novelty filter, (ii) explorative deterministic
noise, (iii) a fundamental form of neural activity that provides
continuous, sequential access to memory patterns, and (iv) a mechanism
that underlies the formation of complex categories. An experiment in the
artificial evolution of modular neural architectures, demonstrates that by
manipulating modular topology and parameters governing local learning
and activation processes, "genetic algorithms" can effectively explore
complex interactive dynamics to construct efficient, modular neural
architectures for pattern categorization tasks. A particularly striking
result is that coupled, oscillatory circuits were installed by the genetic
algorithm, inducing the formation of fractal category boundaries.
Dynamic representations in these networks, can significantly reduce
sequential interference due to overlapping static representations in
learning neural networks.

File 3:
The above two papers, among others, describe a digit recognition
network. A demonstration version of this can be retrieved for PCs (486
DX recommended):

digidemo.zip           (unzip with PKUNZIP; contains several files)

Some documentation is included with the program. One of its features is
retraining on digits that go wrong. Only the wrong digits are retrained,
without catastrophic interference with the other digits.

With questions and remarks, contact either Bart Happel at Leiden
University (happel at rulfsw.leiden.univ) or Jaap Murre at the MRC
Applied Psychology Unit (jaap.murre at mrc-apu.cam.ac.uk).




From LINDBLAD at vana.physto.se  Fri Dec 16 08:44:45 1994
From: LINDBLAD at vana.physto.se (LINDBLAD@vana.physto.se)
Date: Fri, 16 Dec 1994 15:44:45 +0200
Subject: IBM ZISC Paper
Message-ID: <01HKPT2KNC4Y8Y6ZY4@vana.physto.se>

FTP-HOST: archive.cis.ohio-state.edu
FTP-FILE: pub/neuroprose/eide.zisc.ps.Z

        The file eide.zisc.ps. has been placed into the 
Neuroprose repository:
  
"An implementation of the Zero Instruction Set Computer (ZISC036) 
on a PC/ISA-bus card" (14 pages)
A. Eide, Th. Lindblad, C.S. Lindsey, M. Minerskjoeld, G.Sekhniaidze
and G. Szekely

Abstract: The new IBM Zero Instruction Set Computer (ZISC036) chip has
been implemented on a PC/ISA-bus card. The chip has 36 RBF-like neurons.
It is highy parallel and cascadable with on-chip learning. A card with
two ZISC036 chips was built and tested with a noisy character recognition
"benchmark".
 


From LINDBLAD at vana.physto.se  Fri Dec 16 08:46:00 1994
From: LINDBLAD at vana.physto.se (LINDBLAD@vana.physto.se)
Date: Fri, 16 Dec 1994 15:46:00 +0200
Subject: WWW Page on Neural Nets in High Energy Physics
Message-ID: <01HKPT45B9N68Y6ZY4@vana.physto.se>

 	We would like to announce the installation of
the WWW homepage "Neural Networks in High Energy Physics". 
The address is:

       "http://www1.cern.ch/NeuralNets/nnwInHep.html"

	Both hardware and software neural network techniques have
been used in high energy physics. Hardware neural networks are
used for real-time data selection, while software neural networks
are used in off-line analysis to enhance signal to background
discrimination. We include a long reference list of work done 
in these areas, as well as news of recent developments. 
	Of most general interest is the extensive page on 
commercial hardware neural networks. Descriptions are given
of VLSI chips (e.g. the new IBM ZISC chip with RBF neurons),
accelerator boards, and neurocomputers.
                Clark S. Lindsey (lindsey at msia02.msi.se)
                Bruce Denby (denby at pisa.infn.it)
                Thomas Lindblad (lindblad at vana.physto.se)


From ricart at picard.jmb.bah.com  Fri Dec 16 10:32:22 1994
From: ricart at picard.jmb.bah.com (Rick Ricart)
Date: Fri, 16 Dec 94 10:32:22 EST
Subject: Apology for Flame, and Survey
Message-ID: <9412161532.AA03333@picard.jmb.bah.com>


I agree with John Pollock that reviewing fees should not be instituted.
I'm not sure what prompted the unprecedented decision by WCNN, but if one 
reason for the fee is to increase the overall quality of the submissions, 
I offer the following alternative.  Have each society (IEEE, INNS, etc.) 
establish and publish minimum acceptance criteria for papers.  These criteria 
might include, for example, precise algorithm and parameter descriptions so that 
experiments can be reproduced by others.  I know this is a novel idea for 
some current scientific and engineering societies and publishers, but I think
we're ready for it.

The real reason for the reviewing fee might be, of course, financial.  All 
reviewers are very busy individuals with many professional duties.  Some, as is
evident, feel they should be reimbursed for "volunteering" their precious
time for the given society's benefit.  My answer to the problem is, "don't 
volunteer."  There are many other professionals in the field that would 
gladly volunteer their time to review papers given the opportunity; especially
if they have clear minimum acceptance guidelines available from the given
society.

These are my personal feelings and in no way do they represent an official 
Booz-Allen & Hamilton, Inc view point.

Rick Ricart					Associate
Booz-Allen & Hamilton   Advanced Computational Technologies Practice
McLean, VA 22020			Phone: (703) 902-5494
email: ricart at picard.jmb.bah.com


From timxb at faline.bellcore.com  Fri Dec 16 11:26:07 1994
From: timxb at faline.bellcore.com (Timothy X Brown)
Date: Fri, 16 Dec 1994 11:26:07 -0500
Subject: Apology for Flame, and Survey
Message-ID: <199412161626.LAA21111@faline.bellcore.com>

Review fees to cover costs may or may not be justified, but it is a
dangerous precedent, especially when an implicit goal was to limit
entry. Organizers of conferences/journals/grants/positions can only
lose by restricting the submission that they receive. Each of the
"problems" Jordan Pollack mentions can be addressed in slightly
different ways:

>a) filtering out computer-generated articles,
I thought "filtering out" is the quintissential role of the review
process. By having some standards for acceptance, such fluff can be
removed. One of the roles of a conference is to do some of this before
hand so I don't waste valuable time on low information content.

>b) authors not showing up to deliver talks or posters,
Have no review fee, but the authors must submit the conference fee
which is then refundable if the paper is not accepted.

>c) expense of supporting travel costs for both plenary 
   speakers and conference organizers, and
If the organizers cannot organize a cost effective conference, get
someone else to do it. I know that the American Social Sciences
Association runs a huge 28 parallel session conference, and conference
fees last year (this was in downtown Boston) were $35 with a $15
student rate! Of course, banquets (choose from one of 10), etc, were
extra and water was the only refreshment provided, but what is the
point of the conference? BTW, I've organized conferences without my
expenses being covered.

>d) declining membership/subscription revenues.
Orgainize smaller conferences (with <6 parallel sessions), smaller
print runs, etc. If profits are the only motive, put a centerfold in
the magazine and have naked dancers at the banquet.

	Tim

Timothy X Brown
MRE 2E-378
Adaptive Systems Research
Bell Communications Research
445 South Street
Morristown, NJ 07960
Tel: (201) 829-4314
Fax: (201) 829-5888



From singh at psyche.mit.edu  Fri Dec 16 16:13:43 1994
From: singh at psyche.mit.edu (Satinder Singh)
Date: Fri, 16 Dec 94 16:13:43 EST
Subject: Transfer/Interference in control problems...
Message-ID: <9412162113.AA17889@psyche.mit.edu>

	This is another note on the topic of transfer and interference
across problems. 

	I have investigated this issue *a bit* in the reinforcement
learning setting of an agent required to solve a SET of hierarchically
structured control problems in the *same* environment. Reinforcement
learning (RL) algorithms solve control problems by using control
experience to update utility functions. With experience, the utility
function improves and therefore so does the resulting ``greedy''
decision strategy.

	I studied the case of compositionally structured control
tasks, where complex tasks are sequences of simpler tasks. I showed
how a modular RL architecture (that is based on Jacobs, Jordan, Nowlan
and Hinton's mixture of experts architecture) is able to reuse utility
functions learned for simple tasks to quickly construct utility
functions for more complex tasks in the hierarchy. The composition of
a complex task is not available to the agent, but has to be learned. 
The use of ``shaping'' to encourage transfer is also illustrated. 

	Anonymous ftp instructions for the above work follow -- the
filename is singh-MLJ-1.ps.Z (another file of possible interest is
singh-ML92.ps.Z).

	Several other RL researchers have since worked on this topic.
See Dayan and Hinton's paper on ``Feudal Reinforcement Learning'' in
NIPS-5 for more recent work and uptodate references.
================================================================

unix> ftp envy.cs.umass.edu

Name: anonymous
Password: [your ID]
ftp> cd pub/singh
ftp> binary
ftp> get <filename>.ps.Z
ftp> bye

unix> uncompress <filename>.ps.Z
unix> [your command to print PostScript file] <filename>.ps


From N.Sharkey at dcs.shef.ac.uk  Fri Dec 16 12:06:27 1994
From: N.Sharkey at dcs.shef.ac.uk (N.Sharkey@dcs.shef.ac.uk)
Date: Fri, 16 Dec 94 17:06:27 GMT
Subject: About sequential learning (or interference)
In-Reply-To: Stephen Hanson's message of Wed, 14 Dec 1994 15:20:45 -0500 (EST) <YivpCRu1GEMnA0vvJt@tractatus.siemens.com>
Message-ID: <9412161706.AA01599@entropy.dcs.shef.ac.uk>


>of course avoiding "interference" is
>another way of preventing generalization.

>As usual there is a tradeoff here.

Yes there is a trade-off but not with interference as Hetherington has pointed
out.

In fact, with backprop, the way to entirely eliminate interference is
to get a good approximation to the total underlying function that is
being sampled. For example, with an autoencoder memory, if the there
is good extraction of the identity function then there will be no
interference from training on sucessive memory sets (and of course
little need for further training).

The trade-off is between old-new discrimination and generalisation. Definitionally,
as one improves the other collapses.

In the paper cited by Heterington (which is now under journal
submission). We present a formally guaranteed solution to the
interference and discrimination problem (the HARM) model, but it
demands exponentially increasing computational resources. It is really
used to show the problems of other localisation solutions (French,
Murre, Kruschke etc.).

We also report some interesting empirical (simulation) results of this trade-off in a
much shorter paper:

Sharkey, NE, and Sharkey, AJC, (in press) Interference and
Discrimination in Neural Net Memory. In Joe Levy, Dimitrios
Bairaktaris, John Bullinaria and Paul Cairns.(Ed) Connectionist Models
of Memory and Language, UCL press.

If anyone is interested I will mail a postscript copy of the tech report to them.

Sharkey, N.E., & Sharkey, A.J.C. Understanding Catastrophic
  Interference in neural Nets.  Technical Report, Department of Computer
  Science, University of Sheffield, U.K. 

Abstract

A number of recent simulation studies have shown that when feedforward
neural nets are trained, using backpropagation, to memorize sets of
items in sequential blocks and without negative exemplars, severe
retroactive interference or {\em catastrophic forgetting} results.
Both formal analysis and simulation studies are employed here to show
why and under what circumstances such retroactive interference arises.
The conclusion is that, on the one hand, approximations to "ideal"
network geometries can entirely alleviate interference, but at the
cost of a breakdown in discrimination between input patterns that have
been learned and those that have not: {\em catastrophic remembering}.
On the other hand, localized geometries for subfunctions eliminate the
discrimination problem but are easily disrupted by new training sets
and thus cause {\em catastrophic interference}.  The paper concludes
with a Hebbian Autoassociative Recognition Memory (HARM) model which
provides a formally guaranteed solution to the problems of
interference and discrimination. This is then used as a yardstick with
which to evaluate other proposed solutions.

noel

 Noel Sharkey  	   	   
 Professor of Computer Science   
 Department of Computer Science  
 Regent Court                    
 University of Sheffield 	 
 S1 4DP, Sheffield, UK           

 N.Sharkey at dcs.shef.ac.uk

FAX: (0742) 780972



From dyyeung at cs.ust.hk  Sat Dec 17 09:54:27 1994
From: dyyeung at cs.ust.hk (Dr. D.Y. Yeung)
Date: Sat, 17 Dec 94 09:54:27 HKT
Subject: elitism at NIPS
Message-ID: <9412170154.AA17552@cs.ust.hk>

> The folks involved in running NIPS are aware that in the past, some people
> have felt the conference was biased against outsiders.  As the program
> chair for NIPS*94, I want to mention some steps we took to reduce that
> perception:
> 
> * This year we required authors to submit full papers, not just abstracts.
>   That made it harder for "famous" people to get by on their reputation
>   alone, and easier for "non-famous" people to get good papers in.
> 
> * This year we required reviewers to provide real feedback to all authors.
>   In order to make this less painful, we completely redesigned the review
>   form (it's publicy accessible via the NIPS home page) and, for the first
>   time, accepted reviews by email.  Everyone liked getting comments from
>   the reviewers, and authors whose papers were not accepted understood why.
> 
> * We continue to recruit new people for positions in both the organizing
>   and program committees.  It's not the same dozen people year after year.
>   We also have a large reviewer pool:  176 people served as reviewers
>   this year.
> 
> * We tend to bring in "outsiders" as our invited speakers, rather than
>   the usual good old boys.  This year's invited speakers included Francis
>   Crick of the Salk Insititute, Bill Newsome (a neuroscientist from 
>   Stanford), and Malcolm Slaney (a signal processing expert formerly with
>   Apple and now at Interval Research.)  None had been to NIPS before.

Another step that the organizers of NIPS may consider is to have a
"blind" review process, which is an even more positive step to reduce
that perception.  This practice has been used in some other good
conferences too.

Regards,

Dit-Yan Yeung


From raymond at fit.qut.edu.au  Sat Dec 17 20:24:16 1994
From: raymond at fit.qut.edu.au (Raymond Lister)
Date: Sat, 17 Dec 94 20:24:16 EST
Subject: elitism at NIPS
Message-ID: <199412171024.UAA20868@fitmail.fit.qut.edu.au>


> From Dave_Touretzky at cs.cmu.edu Sat Dec 17 14:53:02 1994
> To: Connectionists at cs.cmu.edu
> Subject: elitism at NIPS
> Date: Fri, 16 Dec 94 03:22:31 EST
> 
> ...
> 
> * This year we required authors to submit full papers, not just abstracts.
>   That made it harder for "famous" people to get by on their reputation
>   alone, and easier for "non-famous" people to get good papers in.

The move to full papers was a good change.  It would be better still if the
copies of papers sent to referees did not contain the names and addresses
of authors.

Raymond Lister,
School of Computing Science,
Queensland University of Technology,
AUSTRALIA

Internet: raymond at fitmail.fit.qut.edu.au


From gluck at pavlov.rutgers.edu  Sat Dec 17 10:45:25 1994
From: gluck at pavlov.rutgers.edu (Mark Gluck)
Date: Sat, 17 Dec 94 10:45:25 EST
Subject: Rutgers Grad Program in BEHAVIORAL AND NEURAL SCIENCES
Message-ID: <9412171545.AA01777@james.rutgers.edu>

---------------------------------------------------------------------
 
       Seeking Applications for Fall 1995 for Ph.D. Program in

		    BEHAVIORAL AND NEURAL SCIENCES

		      Rutgers University, Newark

	   Target date for applications is JANUARY 20, 1995

---------------------------------------------------------------------

If you are considering graduate study in Cognitive, Integrative,
Molecular, or Computational Neuroscience, you may be interested in
Rutgers' new interdisciplinary research-oriented graduate program in
Behavioral and Neural Sciences (BNS).

The BNS aims to provide students with a rigorous understanding of the
basic tenets and underpinnings of modern neuroscience.  The program
emphasizes the multidisciplinary nature of this endeavor, and offers
specific research training in Behavioral and Cognitive Neuroscience
and Molecular, Cellular and Systems Neuroscience.  These research
areas represent different but complementary approaches to contemporary
issues in behavioral and molecular neuroscience and can emphasize
either human or animal studies.

The graduate program is offered by two distinct university units: the
newly established Center for Molecular and Behavioral Neuroscience
(CMBN) and the Institute of Animal Behavior (IAB). These two units
work together but each has its own special emphasis.

Research at the CMBN emphasizes integration across levels of analysis
and traditional disciplinary boundaries. The CMBN is one of the
leading places in this country for the study of the neural bases of
behavior and cognition in humans and other animals. Behavioral
research areas include the study of memory, language (both signed and
spoken), motor-control, and vision. Clinically relevant research areas
are the study of the physiological and pharmacological aspects of
schizophrenia, epilepsy and Parkinson's disease and molecular genetics
of reading disorders and well as neuroendocrinology. We have a
computational program for students interested in pursuing
neural-network models as a tool for understanding psychological and
biological issues.  There is also a strong focus on single cell (patch
clamp, intracellular and extracellular) electrophysiology and
multi-unit recording, systems analysis, neuroanatomy and in vivo
microdialysis.

The IAB offers a unified program in psychobiology and ethological
patterns of behavior, with an emphasis on evolution, development and
reproduction, as well as the neurogenesis and recovery of function
from brain damage.


Other Info 
---------- 
At present the CMBN supports up to 40 students with 12-month renewable
assistantships for a period of four years. The curent stipend for
first year students is $12,750; this includes tuition remission and
excellent healthcare benefits. The IAB supported students receive D.S.
Lehrman Fellowships which include a 12-month stipend of approximately
$10,500 for four years and tuition remission.  In addition, the
Johnson & Johnson pharmaceutical company's Foundation has provided
four Excellence Awards which increase students' stipends by $5,000.
Several other fellowships are offered.  More information is available
in our graduate brochure.

The Rutgers-Newark campus (as distinct from the New Brunswick campus),
is 30 minutes outside New York City, and close to other major
university research centers at NYU, Columbia, and Princeton, as well
as major industrial research labs in Northern NJ, including ATT,
Bellcore, Siemens, and NEC.


Faculty Associated With Rutgers Behavioral & Neural Sciences Ph.D. Program
--------------------------------------------------------------------------
	FACULTY - RUTGERS

Elizabeth Abercrombie (Ph.D., Princeton), neurotransmitters and behavior [CMBN]
Colin Beer (Ph.D., Oxford), ethology [IAB]
April Benasich (Ph.D., New York), infant perception and cognition [CMBN]
Ed Bonder (Ph.D., Pennsylvania), cell biology [Biology]
Linda Brzustowicz (M.D.,Ph.D., Columbia), human genetics [CMBN]
Gyorgy Buzsaki (Ph.D., Budapest), systems neuroscience [CMBN]
Mei-Fang Cheng (Ph.D., Bryn Mawr) neuroethology/neurobiology [IAB]
Ian Creese (Ph.D., Cambridge), neuropsychopharmacology [CMBN]
Doina Ganea (Ph.D., Illinois Medical School), molecular immunology [Biology]
Alan Gilchrist (Ph.D., Rutgers), visual perception [Psychology]
Mark Gluck (Ph.D.,Stanford), learning, memory and neural computation [CMBN]
Ron Hart (Ph.D., Michigan), molecular neuroscience [Biology]
G. Miller Jonakait (Ph.D., Cornell Medical College), neuroimmunology [Biology]
Judy Kegl (Ph.D., M.I.T.), linguistics/neurolinguistics [CMBN]
Barry Komisaruk (Ph.D., Rutgers), behavioral neurophysiology/pharmacology [IAB]
Sarah Lenington (Ph.D., Chicago), genetic basis of mating preference [IAB]
Joan Morrell (Ph.D., Rochester), cellular neuroendocrinology [CMBN]
Teresa Perney (Ph.D., Chicago), ion channel gene expression and function [CMBN]
Howard Poizner (Ph.D., Northeastern), language and motor behavior [CMBN]
Jay Rosenblatt (Ph.D., New York), maternal behavior [IAB]
Anne Sereno (Ph.D., Harvard), attention and visual perception [CMBN]
Maggie Shiffrar (Ph.D., Stanford), vision and motion perception[CMBN]
Harold Siegel (Ph.D., Rutgers) neuroendocrine mechanisms [IAB]
Ralph Siegel (Ph.D., McGill), neuropsychology of visual perception [CMBN]
Donald Stein (Ph.D., Oregon), neural plasticity [IAB]
Jennifer Swann (Ph.D., Michigan), neuroendocrinology [Biology]
Paula Tallal (Ph.D., Cambridge), neural basis of language development [CMBN]
James Tepper (Ph.D., Colorado), basal ganglia neurophysiology and anatomy [CMBN]
Beverly Whipple (Ph.D., Rutgers), women's health [Nursing]
Laszlo Zaborszky (Ph.D., Hungarian Academy), neuroanatomy of forebrain [CMBN]

	BNS FACULTY - UMDNJ 
Barry Levin (M.D., Emory Medical) neurobiology
Benjamin Natelson (M.D., Pennsylvania) stress and distress
Allan Siegel (Ph.D., SUNY-Buffalo), aggressive behavior
Walter Tapp (Ph.D., Cornell), primate models of cognitive function
	
        ASSOCIATES OF CMBN
Izrail Gelfand (Ph.D., Moscow State), biology of cells [Biology]
Richard Katz (Ph.D., Bryn Mawr), psychopharmacology [Ciba Geigy]
David Tank (Ph.D., Cornell), neural plasticity [Bell Labs]


For More Information or an Application
--------------------------------------
If you are interested in applying to our graduate program, or possibly
applying to one of the labs as a post-doc, research assistant or
programmer, please contact us via one of the following:

	Dr. Gyorgy Buzsaki or Dr. Mark A. Gluck 
        CMBN, Rutgers University
        197 University Ave.
        Newark, New Jersey  07102

	Phone (Secretary, Ann Kutyla):  (201) 648-1080 (Ext. 3200) 
        Fax:    			(201) 648-1272
	Email:  buzsaki at axon.rutgers.edu or
		gluck at pavlov.rutgers.edu

We will be happy to send you info on our research and graduate
program, as well as set up an a possible visit to the Neuroscience
Center here at Rutgers-Newark.


INTERNET INFORMATION:
---------------------
Additional information on this program can be obtained 
over the internet via World Wide Web at:

	http://www.cmbn.rutgers.edu/

Please be warned that it is still under construction.



From sef+ at cs.cmu.edu  Sat Dec 17 15:20:28 1994
From: sef+ at cs.cmu.edu (Scott E. Fahlman)
Date: Sat, 17 Dec 94 15:20:28 EST
Subject: elitism at NIPS 
In-Reply-To: Your message of Sat, 17 Dec 94 20:24:16 -0500.
             <199412171024.UAA20868@fitmail.fit.qut.edu.au> 
Message-ID: <mailman.723.1149540302.24850.connectionists@cs.cmu.edu>


Personally, I think that "blind" reviewing is a bad idea because it is
dishonest.  In a field like this one, it is easy in 90% of the cases for an
experienced reviewer to tell who is the author of a paper, or at least what
group the paper is from.  I think it's better to be explicit about the lack
of anonymity, and to challenge the reviewers to consciously try to rise
above any "in group" bias than to provide a show of anonymity that is
really a fraud in most cases.

I also believe that in some cases the identity of the author does provide
essential context to the reviewer.  If I were to present evidence that my
own Cascade-Correlation algorithm gets certain things wrong, a reviewer
might not have to worry as much about rookie mistakes in running Cascor as
he would in a paper from someone he has never heard of.  On the other hand,
if I claim that Cascor does certain things *better* than the competition,
the claim might warrant extra scrutiny because I am an interested party in
the debate.  Even more scrutiny is called for if I am known to have a
long-standing feud with the person whose work I am criticizing.

I know that this view is controversial.  Some of you will certainly argue
that such considerations have no place in science, and that every paper
must stand or fall on its content alone.  Perhaps that is true in a long
paper, but it may be impossible to present all the relevant context in an
eight-page paper. (Because of the NIPS style, this is more like a 5-page
paper published elsehwere.)

I don't know if this is the right forum for a debate on the merits of blind
reviewing.  I just thought it would be useful to point out that there are
some arguments on the other side.

-- Scott

===========================================================================
Scott E. Fahlman			Internet:  sef+ at cs.cmu.edu
Principal Research Scientist		Phone:     412 268-2575
School of Computer Science              Fax:       412 268-5576
Carnegie Mellon University		Latitude:  40:26:46 N
5000 Forbes Avenue			Longitude: 79:56:55 W
Pittsburgh, PA 15213			Mood:      :-)
===========================================================================


From peter at ai.iit.nrc.ca  Sun Dec 18 22:41:32 1994
From: peter at ai.iit.nrc.ca (Peter Turney)
Date: Mon, 19 Dec 1994 08:41:32 +0500
Subject: Workshop on Data Engineering for Inductive Learning
Message-ID: <9412191341.AA05163@ksl0j.iit.nrc.ca>




CALL FOR PARTICIPATION: Workshop on Data Engineering for Inductive Learning
---------------------------------------------------------------------------

IJCAI-95, Montreal (Canada), August 19/20/21, 1995


Objective
---------

In inductive learning, algorithms are applied to data. It is
well-understood that attention to both elements is critical -- unless
instances are represented so as to make its generalization methods
appropriate, no inductive learner can succeed. In applied work, it is
not uncommon for practitioners to spend the bulk of their time
exploring and transforming data in efforts to enable the use of
existing induction techniques.

Despite widespread acceptance of these facts, however, research reports
normally give data work short shrift. In fact, a report devoted mainly
to the data in an induction problem rather than to the algorithms that
process it might well be difficult to publish in mainstream machine
learning and neural network venues. 

Our goal in this workshop is to counterbalance the predominant focus on
algorithms by providing a forum in which data takes center stage.
Specifically, we invite discussion of issues relevant to data
engineering, which we define as the transformation of raw data into a
form useful as input to algorithms for inductive learning. Data
engineering is a concern in industrial and commercial applications of
machine learning, neural networks, genetic algorithms, and traditional
statistics. Among others, papers of the following kind are welcome:

1. Detailed case studies of data engineering in real-world applications 
   of inductive learning.

2. Descriptions of data engineering techniques or methods that have
   proven useful across a number of applications.

3. Studies of the data requirements of important inductive learning
   algorithms, the specifications to which data must be engineered for
   these algorithms to function.

4. Reports on software tools and environments for data engineering,
   including papers on "interactive induction" algorithms.

5. Empirical studies documenting the effect of data engineering on the
   success of induced models.

6. Surveys of data engineering practice in related fields: statistics,
   pattern recognition, etc. (but not problem-solving or 
   theorem-proving).

7. Papers on constructive induction, feature selection and related
   techniques.

8. Papers on (re)formulating a problem, to make it suitable
   for inductive learning techniques. For example, a paper
   on reformulating the problem of information filtering as
   learning to classify.

This workshop will enable an overview of current work in data
engineering. Since the problem of data engineering has received
relatively little published attention, it is difficult to anticipate
the work that will be presented at this workshop. We expect that the
workshop will make it possible to see common trends, shared problems,
and clever solutions that we cannot guess at, given our current,
limited view of data engineering. We have allowed ample time for
discussion of each paper (10 minutes), to foster an atmosphere
that will encourage data engineers to share their stories and to seek
common elements. We aim to leave the workshop with a vision of the
research directions that might bring science into data engineering.


Participation
-------------

During the workshop, we anticipate approximately 14 presentations. Each
paper will be given 25 minutes, 15 minutes for presentation and 10
minutes for discussion. There will be at most 30 participants in the
workshop. If you wish to participate in the workshop, you may either
submit a paper or a description of work that you have done (are doing,
plan to do) that is relevant to the workshop. Papers should be at most
10 pages long. The first page should include the title, the author's
name(s) and affiliation(s), a complete mailing address, phone number,
fax number, e-mail, an abstract of at most 300 words, and up to five
keywords. For those who do not choose to submit a paper, a description
of relevant work should be at most 1 page long and should include
complete address information. Workshop participants are required to
register for the main IJCAI-95 conference.

All submissions (papers or descriptions of relevant work) will be
reviewed by at least two members of the organizing committee. Please
send your submissions to the contact address below. Submissions should
be PostScript files, sent by e-mail. Accepted submissions will be
available before the workshop through ftp. Workshop participants will
also be given copies of the papers on the day of the workshop.

In selecting the papers, the committee will aim for breadth of coverage
of the topics listed above. Ideally, each of the eight kinds of papers
listed above would have at least one representative in the workshop.
A paper with new ideas on data engineering will be preferred to a high-
quality paper on a familiar idea.

The workshop organizers plan to publish revised versions of selected
papers from the workshop. The papers would be published either as a
book or as a special issue of a journal.

The exact date for the workshop has not yet been decided by IJCAI.
The workshop is one day in duration and will be held on one of
August 19, 20, or 21.


Schedule
--------

Deadline for submissions:		March 31, 1995
Notification of acceptance:		April 21, 1995
Submissions available by ftp:		April 28, 1995
Actual Workshop:			August 19/20/21, 1995


Organizing Committee
--------------------

Peter Turney, National Research Council (Canada)
Cullen Schaffer, CUNY/Hunter College (USA)
Rob Holte, University of Ottawa (Canada)


Contact Address
---------------

Dr. Peter Turney
Knowledge Systems Laboratory
Institute for Information Technology
National Research Council Canada
Ottawa, Ontario, Canada
K1A 0R6
(613) 993-8564 (office)
(613) 952-7151 (fax)
peter at ai.iit.nrc.ca









From ucganlb at ucl.ac.uk  Mon Dec 19 09:08:28 1994
From: ucganlb at ucl.ac.uk (Dr Neil Burgess - Anatomy UCL London)
Date: Mon, 19 Dec 94 14:08:28 +0000
Subject: catastrophic interference of BP
Message-ID: <187786.9412191408@link-1.ts.bcc.ac.uk>

Studies of catastrophic interference in BP networks are
interesting when considering such a network as a model of some human
(or animal) memory system. 
Is there any reason for doing that?
Neil


From ted at SPENCER.CTAN.YALE.EDU  Mon Dec 19 09:52:27 1994
From: ted at SPENCER.CTAN.YALE.EDU (ted@SPENCER.CTAN.YALE.EDU)
Date: Mon, 19 Dec 1994 09:52:27 -0500
Subject: "Blind" reviews are impossible
Message-ID: <199412191452.AA23787@PLANCK.CTAN.YALE.EDU>

Even a nonexpert reviewer can figure out who wrote a paper simply by
looking for citations of prior work.  The only way to guarantee a
"blind" review is to forbid authors from citing anything they've done
before, or insist on silly euphemisms when citing such publications.

--Ted


From jbower at smaug.bbb.caltech.edu  Mon Dec 19 15:44:10 1994
From: jbower at smaug.bbb.caltech.edu (jbower@smaug.bbb.caltech.edu)
Date: Mon, 19 Dec 94 12:44:10 PST
Subject: WCNN / INNS / or whatever
Message-ID: <9412192044.AA00733@smaug.bbb.caltech.edu>

In brief support of Jordon, I believe that it is well known that the
finances of both the WCNN and the INNS have been strange for years. It is
also not too surprising that one would confuse the two, same list of
participants, basically.  It is also well known that the "everyone who can
pay" approach taken by the organizers of most of the neural network
meetings has resulted in very poor signal to noise ratios.  The
historically more "old boy" approach of NIPS has meant a better quality
meeting, but less openness.  I wish Dave Touretzky luck in efforts to
change, there is little evidence that INNS or IEEE even recognize that
there is a problem.   

Two years ago, a considerable amount of pressure was placed on the
neurobiologists who have organized the computational neuroscience meetings
(CNS*92-94) to merge with the next INNS meeting (in this case).  The vote
by participants of CNS*93 not to, was essentially 100%. Other than the
comment that very little that goes on in any of these meetings has much to
do with neurobiology, the other reasons most often mentioned where the
above.


Jim Bower    

           ***************************************
                       James M. Bower
                     Division of Biology
                     Mail code:  216-76
                           Caltech
                     Pasadena, CA 91125
                      (818) 395-6817
                      (818) 449-0679 FAX

      NCSA Mosaic laboratory address:  http://www.bbb.caltech.edu/bowerlab
      NCSA Mosaic address for GENESIS:  http://www.bbb.caltech.edu/GENESIS




From maass at igi.tu-graz.ac.at  Mon Dec 19 15:48:06 1994
From: maass at igi.tu-graz.ac.at (Wolfgang Maass)
Date: Mon, 19 Dec 94 21:48:06 +0100
Subject: elitism at NIPS 
Message-ID: <199412192048.AA16441@figids01>


I have the impression that for theory-papers it would NOT be 
very benefitial if NIPS would change to "blind reviewing". 


From dtam at morticia.cnns.unt.edu  Mon Dec 19 16:28:15 1994
From: dtam at morticia.cnns.unt.edu (David Tam)
Date: Mon, 19 Dec 94 15:28:15 CST
Subject: elitism at NIPS
Message-ID: <199412192128.AA02352@morticia.cnns.unt.edu>

I think a totally honest system has to be doubly-open, i.e., the reviewers
names have to be attached to the referee's comments.  That will hold them
accountable for what they say.  That will keep them HONEST.  If they want to
give negative critisim, let it be known who they are.

As is now, it is not an open system -- It is blind one way, and open the
other way, and that makes the system so unfair, bias, and dictatorial.

As Scott Fahlman said, it is practically impossible to make it blind,
because we all know who does what line of work.  So, if it is not blind,
make it totally open, and have a repeal process, such that if the
referee is WRONG, there is a recourse!!

That's what a democratic process is all about -- to keep the system
honest by having an OPEN process with REPEAL recourse.  A closed system
is a sure way to breed corruption and dictatorship.

This response applies to the whole scientific review process in reviewing
papers and grants in general as well as in conferences.

David C. Tam
Center for Network Neuroscience
Dept. of Biological Sciences
University of North Texas



From shams at maxwell.hrl.hac.com  Mon Dec 19 20:39:28 1994
From: shams at maxwell.hrl.hac.com (Soheil Shams)
Date: Mon, 19 Dec 1994 17:39:28 -0800
Subject: Position Announcement
Message-ID: <9412200146.AA07164@maelstrom>

SIGNAL / IMAGE PROCESSING RESEARCH OPPORTUNITIES

Hughes Research Laboratories has an immediate opening for a Research Staff
Member to join a team of scientists in the Computational Intelligence and
Advanced Signal Processing Algorithms Project.  Team members in this
project have developed novel, state-of-the-art neural networks,
time-frequency transforms, and image compression algorithms for use in both
commercial and military applications.

The successful candidate will investigate advanced signal and image
processing techniques for multimedia compression, data fusion, and pattern
recognition applications.  Current work is focused on the application of
wavelets, neural networks, and computer vision techniques for data
compression and pattern recognition applications.  Specific duties will
include theoretical analysis, algorithm design, and software simulation.

Candidates are expected to have a Ph.D. in Electrical Engineering, Applied
Mathematics, or Computer Science.  Strong analytical skills and
demonstrated ability to perform creative research, along with experience in
signal and image processing, image or video compression, or neural networks
are required.  Practical experience with C or C++ is essential.   Good
communications and teamwork skills are keys to success.

Overlooking the Pacific Ocean and the coastal community of Malibu, the
Research Laboratories provides an ideal environment for you to make the
most of your scientific abilities.  Our organization offers a competitive
salary and benefits package.  

Additional information may be obtained from Lynn Ross.  For immediate
consideration, send your resume to:

Lynn W. Ross
Department RM
Hughes Research Laboratories
3011 Malibu Canyon Road
Malibu,  CA   90265
FAX: (310) 317-5651
Internet:  lross at msmail4.hac.com

Proof of legal right to work in the United States required.  An Equal
Opportunity Employer.



From wermter at nats2.informatik.uni-hamburg.de  Tue Dec 20 06:28:41 1994
From: wermter at nats2.informatik.uni-hamburg.de (Stefan Wermter)
Date: Tue, 20 Dec 94 12:28:41 +0100
Subject: IJCAI95-workshop: Learning for Natural Language Processing
Message-ID: <9412201128.AA01573@nats2.informatik.uni-hamburg.de>


---------------------------------------------------------------------------






                       CALL FOR PAPERS AND PARTICIPATION


                            IJCAI-95 Workshop on


         New Approaches to Learning for Natural Language Processing


       International Joint Conference on Artificial Intelligence (IJCAI-95)
                      Palais de Congres,  Montreal, Canada

                     currently scheduled for August 21, 1995



ORGANIZING COMMITTEE
--------------------

Stefan Wermter          Gabriele Scheler              Ellen Riloff
University of Hamburg   Technical University Munich   University of Utah


PROGRAM COMMITTEE
-----------------

Jaime Carbonell, Carnegie Mellon University, USA
Joachim Diederich, Queensland University of Technology, Australia
Georg Dorffner, University of Vienna, Austria
Jerry Feldman, ICSI, Berkeley, USA
Walther von Hahn, University of Hamburg, Germany
Aravind Joshi, University of Pennsylvania, USA
Ellen Riloff, University of Utah, USA
Gabriele Scheler, Technical University Munich, Germany
Stefan Wermter, University of Hamburg, Germany


WORKSHOP DESCRIPTION
--------------------

In the last few years, there has been a great deal of interest and activity  in
developing  new approaches to learning for natural language processing. Various
learning methods have been used, including

- connectionist methods/neural networks
- machine learning algorithms
- hybrid symbolic and subsymbolic methods
- statistical techniques
- corpus-based approaches.

In general, learning methods are designed to support  automated  knowledge  ac-
quisition,  fault  tolerance,  plausible  induction, and rule inferences. Using
learning methods for natural language processing is  especially  important  be-
cause  language learning is an enabling technology for many other language pro-
cessing problems, including  noisy speech/language integration, machine  trans-
lation,  and information retrieval. Different methods support language learning
to various degrees but, in general, learning is  important  for  building  more
flexible, scalable, adaptable, and portable natural language systems.

This workshop is of interest particularly at this time because systems built by
learning  methods  have reached a level where they can be applied to real-world
problems in natural language processing and where they  can  be  compared  with
more traditional encoding methods. The workshop will bring together researchers
from the US/Canada, Europe, Japan, Australia and other countries working on new
approaches to language learning.

The workshop will provide a forum for discussing  various  learning  approaches
for  supporting  natural  language processsing. In particular the workshop will
focus on questions like:

- How can we apply suitable existing learning methods for language processing?

- What new learning methods are needed for language processing and why?

- What language knowledge should be learned and why?

- What are similarities and differences between different approaches for
  language learning? (e.g., machine learning algorithms vs neural networks)

- What are strengths and limitations of learning rather than manual encoding?

- How can learning and encoding be combined in symbolic/connectionist systems?

- Which aspects of system architectures and knowledge engineering have to
  be considered? (e.g., modular, integrated, hybrid systems)

- What are successful applications of learning methods in various fields?
  (speech/language integration, machine translation, information retrieval)

- How can we evaluate learning methods using real-world language?
  (text, speech, dialogs, etc.)


WORKSHOP FORMAT
---------------

The workshop will provide a forum for the interactive  exchange  of  ideas  and
knowledge. Approximately 30-40 participants are expected and there will be time
for up to 15 presentations depending on the number and quality of paper contri-
butions received. Normal presentation length will be 15+5 minutes, leaving time
for direct questions after each talk. There may be a few invited talks of  25+5
minutes length.  In addition to prepared talks, there will be time for moderat-
ed discussions after two related sessions. Furthermore, the  moderated  discus-
sions  will provide an opportunity for an open exchange of comments, questions,
reactions, and opinions.

PUBLICATION
-----------

Workshop proceedings will be published by AAAI.  If  there  is  sufficient  in-
terest  of  the participants of the workshop there may be a possibility to pub-
lish the results of the workshop as a book.

REGISTRATION
------------

This workshop will take place directly before the general IJCAI-conference.  It
is  an IJCAI policy, that workshop participation is not possible without regis-
tration for the general conference.

SUBMISSIONS
-----------

All submissions will be refereed by the program committee and other experts  in
the  field.  Please submit 4 hardcopies AND a postscript file. The paper format
is the IJCAI95 format:  12pt article style latex, no more  than  43  lines,  15
pages  maximum,  including title, address and email address, abstract, figures,
references. Papers should fit to 8 1/2" x 11" size. Notifications will be  sent
by email to the first author.

Postscript files can be uploaded with anonymous ftp:

ftp nats4.informatik.uni-hamburg.de (134.100.10.104)
login: anonymous
password: <your email address>
cd incoming/ijcai95-workshop
binary
put <yourfile.Z or yourfile.gz>
quit

Hardcopies AND postscript files  must arrive not later than 24th February  1995
at the address below.

##############Submission Deadline:  24th February 1995
##############Notification Date: 24th March 1995
##############Camera ready Copy: 13th April 1995



Please send correspondence and submissions to:


################################################
Dr. Stefan Wermter
Department of Computer Science
University of Hamburg
Vogt-Koelln-Strasse 30
D-22527 Hamburg
Germany

phone:  +49 40 54715-531
fax:    +49 40 54715-515
e-mail: wermter at informatik.uni-hamburg.de
################################################


































From N.Sharkey at dcs.shef.ac.uk  Tue Dec 20 08:04:01 1994
From: N.Sharkey at dcs.shef.ac.uk (N.Sharkey@dcs.shef.ac.uk)
Date: Tue, 20 Dec 94 13:04:01 GMT
Subject: catastrophic interference of BP
In-Reply-To: Dr Neil Burgess - Anatomy UCL London's message of Mon, 19 Dec 94 14:08:28 +0000 <187786.9412191408@link-1.ts.bcc.ac.uk>
Message-ID: <9412201304.AA18623@entropy.dcs.shef.ac.uk>


>Studies of catastrophic interference in BP networks are
>interesting when considering such a network as a model of some human
>(or animal) memory system. 
>Is there any reason for doing that?
>Neil

BP has been used extensively in human cognitive and memory modelling and
so it is useful to know if "cashes in" in terms of recognition memory as
well.

The interference issue is also problematic for training online control processes
(e.g. in robotic coordination) where new training data is not necessarily accompanied
by all of the "old" data samples. If there is not a nice regular general function to
be extracted, then forgetting could be very severe. Problems also arise for BP in any
such online task that requires discrimination between what was in the training sets and
what was not.
noel



From gbugmann at school-of-computing.plymouth.ac.uk  Tue Dec 20 12:15:05 1994
From: gbugmann at school-of-computing.plymouth.ac.uk (Guido.Bugmann xtn 2566)
Date: Tue, 20 Dec 94 17:15:05 GMT
Subject: PhD Position available
Message-ID: <210.9412201715@subnode.soc.plym.ac.uk>


A PhD position is available immediately to work on the project "The
Plymouth Hand" at the University of Plymouth, UK.  This project comprises
an application of neural networks to control.

The aim of that project is to develop a robot/prosthetic hand with slippage
detection as a source of feedback for the control of the grip pressure.

A working prototype has been built which can lift a cylinder of solid steel
and also a hollow rolled sheet of A4 paper without deforming it.

Further work comprises:

1. Derivation of a practical specification for two systems using the
slippage detection adaptive grasping force technique, i.e. a robot gripper
and a prosthetic hand.

2. Kinematically model a range of proposed designs using virtual reality
software.

3. Validate appropriate kinematic models

4. Design, build and test an improved version of the existing device to
meet the specifications of a robot gripper. In particular:

a) Software environment will be investigated e.g.  assembly code, C, etc
and an appropriate one chosen and developed. New techniques such as rule
based (fuzzy logic) and neural networks will be investigated.

b) Design and development of the power electronics and drive system.

c) Specify and test the necessary microcontroller.

5) Same as 4) but as an adaptation of an existing NHS prostethic hand.
----------------------------------------------------

The salary in the standard range 5-6 Kpounds/year
for PhD students.

For further information, please
contact:

Paul Robinson
Dept. of Electrical and Electronic Engineering
University of Plymouth
Plymouth PL4 8AA
United Kingdom

Phone (+44) 1752 23 25 95 / 72
Fax (+44) 1752 23 25 83
-----------------------------------------------------


From ucganlb at ucl.ac.uk  Tue Dec 20 12:54:01 1994
From: ucganlb at ucl.ac.uk (Dr Neil Burgess - Anatomy UCL London)
Date: Tue, 20 Dec 94 17:54:01 +0000
Subject: catastrophic interference of BP 
Message-ID: <154923.9412201754@link-1.ts.bcc.ac.uk>

>> Studies of catastrophic interference in BP networks are
>> interesting when considering such a network as a model of some human
>> (or animal) memory system. 
>> Is there any reason for doing that?
>> Neil

> So far Jay McClelland has replied: his recently advertised 
> tech. report provides an example of useful consideration of cat. int. with repect to
> the possible existence of complimentary learning systems in the brain, and attempts 
> to distinguish between the specific properties of BP and the more general question.
> Japp Murre also replied that he partially addresses the problem in [1].

    Generally, I think that caution should be expressed in generalising
between artificial learning mechanisms. It may be that learning within a
system with fixed parameters, mediated by iterative minimisation of some
global attribute (e.g. sum-squared error) will tend to show interference
with `catastrophic' characteristics, although how it manifests itself will
depend on the details of the algorithm.
    But I would be surprised if other algorithms, e.g.  learning by
piecemeal constructive algorithms (where extra units are added for a
specific local task, such as [2]) behaved like that - so e.g. we might not
expect learning like song-learning in birds (associated with neural growth)
to necessarily show the same type of interference.
    I also suspect that the characteristics of interference differ between
iterative algorithms in which the `training set' must be presented many
times (e.g. BP) and `one-shot' learning algorithms (e.g. Hopfield model).
In the former case there is obviously a problem of how to interleave new
data into the training set, in the latter case there is no such problem,
and slight variations of the `Hebbian' learning rule can produce
imprinting, primacy, recency or combinations of the above [3].
    Given the availiblity of alternatives, it is not clear that BP should
always be the canonical choice for modelling learing and memory. It is
certainly not the easiest to motivate biologically.

Merry Christmas,

Neil

[1] report:  ftp://ftp.mrc-apu.cam.ac.uk/pub/nn/murre/hyper1.ps
[2] M. R. Frean, `The Upstart algorithm: A method for
  constructing and training feedforward neural networks',
  {\it Neural Computation,} {\bf 2}, 198-209 (1990).
[3] `List Learning in Neural Networks', Neil Burgess, J. L. Shapiro
  and M. A. Moore, {\it Network} {\bf 2} 399-422, 1991.



From nburgess at lbs.lon.ac.uk  Tue Dec 20 11:49:43 1994
From: nburgess at lbs.lon.ac.uk (nburgess@lbs.lon.ac.uk)
Date: Tue, 20 Dec 94 16:49:43 GMT
Subject: NNCM 95 Announcement
Message-ID: <199412201649.LAA04285@jupiter.lbs.lon.ac.uk>


                      ANNOUNCEMENT AND CALL FOR PAPERS

                                  NNCM-95        
                                    

                      THIRD INTERNATIONAL CONFERENCE ON
                    NEURAL NETWORKS IN THE CAPITAL MARKETS


                    Thursday-Friday, October 12-13, 1995
                with tutorials on Wednesday, October 11, 1995.

                    The Langham Hilton, London, England.



        Neural networks are now emerging as a major modelling methodology 
        in financial engineering. Because of the overwhelming interest of 
        the NNCM workshops held in London in 1993 and Pasadena in 1994, 
        the third annual NNCM conference will be held October 12-13, 1995, 
        in London. NNCM'95 takes a critical look at state of the art neural 
        network applications in finance.

        This is a research meeting where original, high-quality contribu-
        tions to the field are presented and discussed. In addition, a
        day of introductory tutorials (Wednesday, October 11) will be
        included to familiarise audiences of different backgrounds with
        financial engineering, and the mathematical aspects of the field.


        Application areas include:

        +     Bond and stock valuation and trading
        +     Foreign exchange rate prediction and trading
        +     Commodity price forecasting
        +     Risk management
        +     Tactical asset allocation
        +     Portfolio management
        +     Option Pricing
        +     Trading strategies


        Technical areas include but not limited to:

        +     Generalised least squares
        +     Robust model estimation
        +     Univariate time series analysis
        +     Multivariate data analysis
        +     Classification and ranking
        +     Pattern recognition
        +     Model selection
        +     Hypothesis testing and confidence intervals

        Instructions for Authors

        Authors who wish to present a paper should mail a copy of their 
        extended abstract (4 pages, single-sided, single-spaced) typed on 
        A4 (8.5" by 11") paper to the secretariat no later than May 31, 
        1995. Submissions will be refereed by no less than four referees 
        and authors will be notified on acceptance by 30 July 1995. Sep-
        arate registration is required using the attached registration 
        form. Authors are encouraged to submit abstracts as soon as pos-
        sible.

        Registration

        To register, complete the registration form and mail to the  sec-
        retariat. Please note that places are limited and will be allocated
        on a "first-come first-served" basis.

        Secretariat:

        For further information, please contact the NNCM-95 secretariat:

                Ms Busola Oguntula, London Business School
                Sussex Place, Regent's Park, London NW1 4SA, UK
                e-mail:  boguntula at lbs.lon.ac.uk
                phone (+44) (0171) 262 50 50 
                fax (+44) (0171) 724 78 75

        Location:

        The main conference will be held at The Langham Hilton which is
        situated near Regent's Park and is a short walk from Baker Street
        Underground Station. Further directions including a map will be
        sent to all registries.

        Programme Commitee

        Dr A. Refenes, London Business School (Chairman)
        Dr Y. Abu-Mostafa, Caltech
        Dr A. Atiya, Cairo University
        Dr N. Biggs, London School of Economics
        Dr D. Bunn, London Business School
        Dr M. Jabri, University of Sydney
        Dr B. LeBaron, University of Wisconsin
        Dr A. Lo, MIT Sloan School
        Dr J. Moody, Oregon Graduate Institute
        Dr C. Pedreira, Catholic University PUC-Rio
        Dr M. Steiner, Universitaet Munster
        Dr A. Timermann, University of California, San Diego
        Dr A. Weigend, University of Colorado
        Dr H. White, University of California, San Diego



        Hotel Accommodation:

        Convenient hotels include:

                The Langham Hilton
                1 Portland Place
                London W1N 4JA
                Tel: (+44) (0171) 636 10 00 
                Fax: (+44) (0171) 323 23 40

                Sherlock Holmes Hotel
                108 Baker Street, London NW1 1LB
                Tel: (+44) (0171) 486 61 61 
                Fax: (+44) (0171) 486 08 84

                The White House Hotel
                Albany St., Regent's Park, London NW1
                Tel: (+44) (0171) 387 12 00 
                Fax: (+44) (0171) 388 00 91


--------------------------Registration Form ----------------------------

                        NNCM-95 Registration Form

                    Third International Conference on
                  Neural Networks in the Capital Markets
                          October 12-13 1995

        Name:____________________________________________________

        Affiliation:_____________________________________________

        Mailing Address: ________________________________________

        _________________________________________________________

        Telephone:_______________________________________________

  ****Please circle the applicable fees and write the total below****


        Main Conference (October 12-13):
                                                         (British Pounds)
                Registration fee                                450                          
                
                Discounted fee for academicians                 250
                (letter on university letterhead required)      

                Discounted fee for full-time students           100
                (letter from registrar or faculty advisor
                required)                                               

        Tutorials (October 11):

        You must be registered for the main conference in order to 
        register for the tutorials.

                                                         (British Pounds)
                Morning Session Only                            100

                Afternoon Session Only                          100

                Both Sessions                                   150

                Full-time students                               50
                (letter from registrar or faculty advisor 
                required)

        TOTAL:  _________                                

        Payment may be made by: (please tick)
        ____ Check payable to London Business School
        ____ VISA         ____Access         ____American Express

        Card Number:___________________________________




--


From pah at unixg.ubc.ca  Tue Dec 20 14:39:18 1994
From: pah at unixg.ubc.ca (Phil A. Hetherington)
Date: Tue, 20 Dec 1994 11:39:18 -0800 (PST)
Subject: catastrophic interference of BP
In-Reply-To: <187786.9412191408@link-1.ts.bcc.ac.uk>
Message-ID: <Pine.SUN.3.91.941220105036.16867B-100000@unixg.ubc.ca>


Neil Burgess wrote:

> Studies of catastrophic interference in BP networks are
> interesting when considering such a network as a model of some human
> (or animal) memory system. 
> Is there any reason for doing that?

Of course.  Network models have properties such as distributed
representations, generalization, interference, content addressability,
etc., that are also true of animal and human memory.  They provide an
alternative framework for construing memory processes that is superior to
box and arrow modeling primarily because they can be 'lesioned' as in
animal experiments or as found in human patients and because they are
'executable'.  Because they are executable (i.e., perform a function) the
effects of these lesions and other parameter manipulations on the
performance of the network can be observed.  There are now many
alternative supervised learning algorithms available, but there are still 
many reasons to continue to study this one. It is most certainly
true that both humans and animals learn via feedback provided by the
effect of erroneous behavior--a process analagous to back prop. 
Unsupervised learning algorithms such as competetive learning do not give
you this.  Unsupervised learning algorithms are not easily executable--you
can't get them to do many simple 'behaviors', like plot trajectories from
starting locations to multiple goals, as in Neil's models.  Of course, any
supervised learning algorithm will confer the ability to train the net to
perform, but there are a couple, mostly pragmatic, reasons to stick with
back prop for now.  Primarily because of the availablity of the
McClelland & Rumelhart books and program disks, back prop is most easily
available and most commonly used.  Back prop already provides the engine
in countless models already published.  Gaining an understanding of the
behavior of this algorithm will enable a better understanding of the flaws
of these already published models.  Thus, its not so much that you *would
want to* use the algorithm, it is that it has already been used.  Lets
understand why we are discarding it before we do so. 

Phil Hetherington
pah at unixg.ubc.ca


From minton at ptolemy-ethernet.arc.nasa.gov  Tue Dec 20 17:39:00 1994
From: minton at ptolemy-ethernet.arc.nasa.gov (Steve Minton)
Date: Tue, 20 Dec 94 14:39:00 PST
Subject: Learning Article
Message-ID: <9412202239.AA09677@ptolemy.arc.nasa.gov>

Readers of this group may be interested in the following article, which
was just published in the Journal of Artificial Intelligence Research
(a journal which is available both online and in print).

Buntine, W.L. (1994)
  "Operations for Learning with Graphical Models", Volume 2, pages 159-225
   Postscript: volume2/buntine94a.ps (1.53M)
               compressed, volume2/buntine94a.ps.Z (568K)

  Abstract: This paper is a multidisciplinary review of empirical,
  statistical learning from a graphical model perspective.  Well-known
  examples of graphical models include Bayesian networks, directed
  graphs representing a Markov chain, and undirected networks
  representing a Markov field.  These graphical models are extended to
  model data analysis and empirical learning using the notation of
  plates.  Graphical operations for simplifying and manipulating a
  problem are provided including decomposition, differentiation, and the
  manipulation of probability models from the exponential family.  Two
  standard algorithm schemas for learning are reviewed in a graphical
  framework: Gibbs sampling and the expectation maximization algorithm.
  Using these operations and schemas, some popular algorithms can be
  synthesized from their graphical specification.  This includes
  versions of linear regression, techniques for feed-forward networks,
  and learning Gaussian and discrete Bayesian networks from data.  The
  paper concludes by sketching some implications for data analysis and
  summarizing how some popular algorithms fall within the framework
  presented.
  
  The main original contributions here are the decomposition techniques
  and the demonstration that graphical models provide a framework for
  understanding and developing complex learning algorithms.


The PostScript file is available via:
   
 -- comp.ai.jair.papers

 -- World Wide Web: The URL for our World Wide Web server is
       http://www.cs.washington.edu/research/jair/home.html

 -- Anonymous FTP from either of the two sites below:
      CMU:   p.gp.cs.cmu.edu        directory: /usr/jair/pub/volume2
      Genoa: ftp.mrg.dist.unige.it  directory:  pub/jair/pub/volume2

 -- automated email. Send mail to jair at cs.cmu.edu or jair at ftp.mrg.dist.unige.it
    with the subject AUTORESPOND, and the body GET VOLUME2/BUNTINE94A.PS
    (either upper or lowercase is fine). 
    Note: Your mailer might find this file too large to handle.
          (The compressed version of this paper cannot be mailed.)

 -- JAIR Gopher server: At p.gp.cs.cmu.edu, port 70. 

For more information about JAIR, check out our WWW or FTP sites, or
send electronic mail to jair at cs.cmu.edu with the subject AUTORESPOND
and the message body HELP, or contact jair-ed at ptolemy.arc.nasa.gov.




From gary at cs.ucsd.edu  Tue Dec 20 18:29:18 1994
From: gary at cs.ucsd.edu (Gary Cottrell)
Date: Tue, 20 Dec 94 15:29:18 -0800
Subject: "Blind" reviews are impossible
Message-ID: <9412202329.AA18047@desi>


Right, silly euphemisms work, or simply saying, "prior work on this
by cottrell, etc." The year I was on the NIPS PC, I was also on
the PC for ACL and AAAI. ACL used blind reviewing. I was sure I was
reviewing a paper by Ken Church, and I was wrong. There was only one
paper of the 15 or so I reviewed that I had any idea who it was.

gary cottrell


From pf2 at st-andrews.ac.uk  Tue Dec 20 17:07:07 1994
From: pf2 at st-andrews.ac.uk (Peter Foldiak)
Date: Tue, 20 Dec 94 22:07:07 GMT
Subject: research assistant job - computational neuroscience
Message-ID: <8312.9412202207@psych.st-andrews.ac.uk>

     University of St Andrews, School of Psychology
     RESEARCH ASSISTANT
     Computational Neuroscience

A one-year research assistantship is available for research
on a novel application of computational methods to the
neurophysiology of the primate visual system with
Dr Peter Foldiak, in collaboration with Dr David Perrett.

The assistant will be involved in the development of software and
experimental procedures. Training in computer programming (C, UNIX),
and preferably in some of the following areas is preferred:
mathematics (optimisation), neural networks, genetic algorithms,
computer graphics (on SGI), vision- or neuroscience.

Salary will be at the appropriate point on the 1B scale for
Research Staff (GBP 13941-17813).
Application forms and further particulars are available from
Personnel Services, University of St Andrews, KY16 9AJ, U.K.
tel: +44 334 462567, (out of hours +44 334 462571) or by fax:
+44 334 462570, to whom completed forms accompanied by a letter
of application should be returned to arrive not later than
Monday, 30 January 1995. Please quote reference number: SH/APS0838.

The University operates an Equal Opportunities Policy.


From jaap.murre at mrc-apu.cam.ac.uk  Wed Dec 21 12:50:03 1994
From: jaap.murre at mrc-apu.cam.ac.uk (Jaap Murre)
Date: Wed, 21 Dec 94 17:50:03 GMT
Subject: sequential inteference
Message-ID: <9412211750.AA26645@rigel.mrc-apu.cam.ac.uk>


In response to recent comments by Neil Burgess, Bob French, Phil
Hetherington, Noel Sharkey, Jay McClelland and others on 'catastrophic
interference', I think it is important to establish that 'vanilla
backpropagation' has by now been eliminated as a valid model of human
learning and memory, its implausible learning transfer being the most
obvious failure. An important reason for this failure can be found in the
nature of the hidden layer representations.
      Catastrophic interference and hypertransfer (i.e., excessive positive
transfer, see Murre, in press a) are both sides of the same coin: hidden-
layer representations in backpropagation bear little relationship to
orthogonalities of input patterns. In the case of interference experiments in
humans the following result is well established: If the input patterns
(stimuli) in two consecutive learning sets A and B are different than there
will be neither interference nor positive transfer (e.g., Osgood, 1949). This
is not the case in backpropagation: orthogonality of the stimuli has no
effect on the reduction of interference. The reason for this is that the
hidden-layer representations are always about equally overlapping *no
matter what the input stimuli are*. Interference and transfer in
backpropagation are only psychologically implausible with respect to its
indifference towards the structure of the input stimuli. Surprisingly enough,
two-layer networks (i.e., with a normal delta-rule, with one layer of
weights) do not suffer from this problem and are in fact well able to model
human interference (see Murre, in press a).
      Having said this, it is also important to consider the various variant
models of error-correcting learning that have been developed recently and
that use backpropagation as a starting point. These models have very
plausible characteristics with respect to learning and categorization in
humans: Kruschke (1990), Gluck (1991), Gluck and Bower (1988, 1990),
Nosofsky, Kruschke, and McKinley (1992), Shanks and Gluck (1994). In
addition, there are now many variant models of backpropagation that do
not suffer from catastrophic interference. These have already been
mentioned in this discussion, so I will not repeat them here.
      From a biological point of view backpropagation is certainly
implausible but it has been show useful in inferring biological plausible
parameters (e.g., Lockery et al., 1989; Zipser and Anderson, 1988).

I can think of several reasons why backpropagation continues to attract
researchers:

1. It is easy to understand.

2. Many simulators are available (see Murre, in press b).

3. It has been shown to approximate all 'well-behaved' functions (e.g.,
Hornik, Stinchcombe, and White, 1989) and thus is often felt to qualify as
a generic learning mechanism. In particular, it can learn non-linearly
separable pattern sets.

4. It has only a few parameters and these are not very critical for the final
results.

5. It possesses most of the basic elements of a 'prototypical' neural
network: distributed representations, graceful degradation, pattern
completion, and adequate generalization of learned behavior. 

I do not myself think that these reasons form necessarily a sufficient
motivation for using backpropagation, and I indeed prefer to work on other
types of learning methods and neural networks. Perhaps, by investigating
the limitations of backpropagation, necessary minimal improvements may
become clear, so that we can replace it - in its leading role - by either a
more plausible variant algorithm, or by a completely different learning
method. 

Merry Christmas,

                      -- Jaap Murre



References

Gluck, M.A. (1991). Stimulus generalization and representation in adaptive
      network models of category learning. Psychological Science, 2, 50-
      55.
Gluck, M.A., & G.H. Bower (1988). From conditioning to category
      learning: an adaptive network model. Journal of Experimental
      Psychology: General, 117, 227-247.
Gluck, M.A., & G.H. Bower (1990). Component and pattern information in
      adaptive networks, Journal of Experimental Psychology: General,
      119, 105-109.
Kruschke, J.K. (1990). ALCOVE: an exemplar-based connectionist model
      of category learning. Psychological Review, 99, 22-44.
Lockery, S.R., G. Wittenberg, W.B. Kristan, Jr., & Garrison W. Cottrell
      (1989). Function of identified interneurons in the leech elucidated
      using neural networks trained by back-propagation. Nature, 340, 468-
      471.
Murre, J.M.J. (in press a). Transfer of learning in backpropagation and in
      related neural network models. In: J. Levy, D. Bairaktaris, J.
      Bullinaria, & P. Cairns (Eds.), Connectionist Models of Memory and
      Language. London: UCL Press. (In our ftp site:
      ftp://ftp.mrc-apu.cam.ac.uk/pub/nn/murre/hyper1.ps) 
Murre, J.M.J., (in press b). Neurosimulators. In: M.A. Arbib (Ed.),
      Handbook of Brain Research and Neural Networks, Cambridge, MA:
      MIT Press. (In our ftp site:
      ftp://ftp.mrc-apu.cam.ac.uk/pub/nn/murre/neurosim1.ps).
Nosofsky, R.M., J.K. Kruschke, and S. C. McKinley (1992). Combining
      exemplar-based category representations and connectionist learning
      rules. Journal of Experimental Psychology: Learning, Memory, and
      Cognition, 18, 211-233.
Osgood, C.E. (1949). The similarity paradox in human learning: a
      resolution. Psychological Review, 56, 132-143.
Shanks, D.R., & M.A. Gluck (1994). Tests of an adaptive network model
      for the identification and categorization of continuous-dimension
      stimuli. Connection Science, 6, 59-89.
Zipser, D., & R.A. Anderson (1988). A back-propagation programmed
      network that simulates response properties of a subset of posterior
      parietal neurons. Nature, 331, 679-684.



From jras at uned.es  Wed Dec 21 13:12:00 1994
From: jras at uned.es (Jose Ramon Alvarez Sanchez)
Date: 21 Dec 94 19:12 +0100
Subject: CFP: W.S. McCulloch, 25 years in Memoriam
Message-ID: <201*jras@uned.es>


                      Preliminary Call For Papers

    CENTRO INTERNACIONAL DE INVESTIGACION EN CIENCIAS DE LA COMPUTACION
                UNIVERSIDAD DE LAS PALMAS DE GRAN CANARIA


                  W.S. McCulloch: 25 years in Memoriam

      INTERNATIONAL CONFERENCE ON BRAIN PROCESSES, THEORIES AND MODELS

             Las Palmas, Gran Canaria, Spain. Nov 12-17, 1995.


Conference Chairman: Moreno-Diaz, R. (Spain)
Org. Comm. Chairman: Mira-Mira, J. (Spain)


Scientific Advisory Comm. (preliminary):

Amari SI (Japan), Anderson J (USA), Arbib M (USA), Beer S (Canada),
Belmonte C (Spain), Blum M (USA), Braitenberg V (Germany), Cull P (USA),
DaFonseca JL (Portugal), Eckorn R (Germany), Harth E (USA),
Herault J (France), Jain LC (Australia), Kauffman S (USA), Kilmer W (USA),
Leibovic KN (USA), Lettvin J (USA), Malsburg C. von der (Germany),
Maturana H(Chile), McCleland J (USA), Mira-Mira J (Spain), Papert S (USA),
Pichler F (Austria), Ricciardi L (Italy), Sato S (Japan),  
Vitozz E (Switzerland), Von Foester H (USA).

Organizing Comm.:

Alvarez JR (Spain), Cabestany J (Spain), Delgado A (Spain), 
Krasner J (USA), Moreno-Diaz jr R (Spain), Pierce A (USA),
Prieto A (Spain), Sanchez JV (Spain), Suarez-Araujo C (Spain).

Organization Staff:

Alonso-Garcia MT (Spain), Perez Ruiz M (Spain).


                             CALL FOR PAPERS

     The conference will consist in a series of invited lectures by leading
scientists icluding M. Arbib, H. Maturana, J. Lettvin and S. Papert related to
the life and work of W.S.McCulloch, an open forum of paper sessions on the
topics listed below and a workshop on the "Embodiments of Mind at the end of
the Century".

Relevant topics include:

      Anatomical, Physiological, Biochemical and Biophysical levels.
      Natural and artificial neural networks.
      Mathematics, systems theory and global properties of the nervous system.
      Conceptual and formal tools in brain function modelling.
      Hybrid systems: symbolic-connectionist links.
      Plasticity, reliability, learning and memory.
      Implications of the work of WS McCulloch for philosophy and psychology.
      Reverse engineering and neurophysiology.

    
     All accepted papers will be published by the MIT Press in the 
pre-conferece proceedings. Those authors will be sent specific instructions
and guidelines. 

     Invited lectures and selected papers will be published by The MIT Press
in a post-conference book.

     Three copies of the intended papers should be sent, prepared according
to the following instructions for authors no later than May 30, 1995 to:

     Prof. Jose Mira-Mira
     Dpto. Informatica y Automatica-UNED
     Senda del Rey s/n
     28040 MADRID-SPAIN

     Voice: +34 (1) 398 7155
     Fax: +34 (1) 398 6697
     e-mail: <jose.mira at uned.es>

     Please include your fax, e-mail, telephone for quick contact
and notification of acceptance.


INSTRUCTIONS FOR AUTHORS

     Authors should submit three copies of full intended papers, not exceeding
eight pages of DIN A4 or 8.5 by 11 inch paper, including figures, tables and
references, in English.

     The centered heading must include:
	
         Title in capitals.
         Name(s) of author(s).
         Address(es) of author(s)
         A 10 lines abstract.

     Three blank lines should be left between each of the above items and four
between the heading and the body of the paper, 1.6 cm left, right, top and
bottom margins, single-spaced and not exceeding the 8-page limit.

     One additional DIN A-4 must be enclosed with the following organizational
data:

     Title and author(s) name(s).
     A list of five keywords.
     A reference to the Topics the paper relates to.
     Postal address, phone, fax and e-mail if available.

	
     All accepted papers will be published in the conference proceedings.


IMPORTANT DATES

      Final call and additonal info: March 1995
      Final date of submission: May 30, 1995
      Notification of acceptance: July 1, 1995
      Conference: November 12-17, 1995


REGISTRATION FORM

------------------------------------------------------------------------
Last Name:
First name
Organization/University:
Address:
Phone:
Fax:
E-mail:

Do you plan to submit a paper:
Tentative title:

Payment:
     Before April 30:
       - Bank draft for 50.000 pts payable on a Spanish Bank to
         FUNDACION UNIVERSITARIA DE LAS PALMAS: MCCULLOCH 95.
       - Money Transfer of 50.000 pts to BEX account No.
         0104 0336 75 0307007836
     After April 30:
       - Late registration fee of 60.000 pts at the conference desk.
------------------------------------------------------------------------
     Please send this registration form to:

     Mrs. Maria Teresa Alonso-Garcia
     CIICC-Universidad de Las Palmas
     Campus de Tafira-Edf. de Informatica
     35017 Las Palmas, SPAIN




From tetewsky at lima.psych.mcgill.ca  Wed Dec 21 15:30:55 1994
From: tetewsky at lima.psych.mcgill.ca (Sheldon Tetewsky)
Date: Wed, 21 Dec 1994 15:30:55 -0500
Subject: catastrophic interference
Message-ID: <199412212030.PAA03615@lima.psych.mcgill.ca>

Neil Burgess recently wrote that
>studies of catastrophic interference in BP networks are interesting when
>considering such a network as a model of some human (or animal) memory
>system.
However, he also questioned whether or not there was
>..any reason for doing that.

In the Tetewsky, Shultz, and Buckingham study ("Assessing interference and
savings in connectionist models of a sequential recognition memory task"),
referenced in Bob French's recent posting, we did some simulations and 
experiments indicating that neural networks can provide a good account of
human performance in a simple recognition memory task when memory is
assessed in terms of savings scores.  The paper is in progress and will
be announced soon. In the meantime, I'll just mention a few relevant
details.

Our work was motivated by Scott Fahlman's idea that his cascade-correlation
algorithm (CC) has certain inherent design features that should give it an
advantage over BP when it comes to dealing with the problem of catastrophic
interference.  We tested this idea by using an encoder version of CC to
model recognition memory.  Results indicated that in contrast to previous
findings, retroactive interference is not that serious a problem for BP
when memory is measured in terms of savings scores. However, CC also
produced a significant increase in savings, relative to BP.  Aside from
this difference in the magnitude of savings, we also have evidence that CC
was better than BP at accounting for the relative number of trials that
subjects used in the different phases of learning.

-- Sheldon Tetewsky 


From rsun at cs.ua.edu  Wed Dec 21 17:05:13 1994
From: rsun at cs.ua.edu (Ron Sun)
Date: Wed, 21 Dec 1994 16:05:13 -0600
Subject: No subject
Message-ID: <9412212205.AA29941@athos.cs.ua.edu>


=========================================================================
                 Call For Papers and Participation

                      The IJCAI Workshop on
                   Connectionist-Symbolic Integration: 
                      From Unified to Hybrid Approaches



                        to be held at IJCAI'95 
                          Montreal, Canada
                          August 19-20, 1995
-------------------------------------------------------------------------



There has been a considerable amount of research in integrating
connectionist and symbolic processing.  While such an approach has
clear advantages, it also encounters serious difficulties and
challenges.  Therefore, various models and ideas have been proposed to
address various problems and aspects in this integration. There is a
growing interest from many segments of the AI community, ranging from
expert systems, to cognitive modeling, to logical reasoning.

Two major trends can be identified in the state of the art: these are
the unified or purely and the hybrid approaches to integration.
Whereas the purely connectionist ("connectionist-to-the-top") approach
claims that complex symbol processing functionalities can be achieved
via neural networks alone, the hybrid approach is premised on the
complementarity of the two paradigms and aims at their synergistic
combination in systems comprising both neural and symbolic components.
In fact, these trends can be viewed as two ends of an entire spectrum.

Up till now, overall, there is still relatively little work in
comparing and combining these fairly isolated efforts. This workshop
will provide a forum for discussions and exchanges of ideas in this
area, to foster cooperative work.  The workshop will tackle important
issues in integrating connectionist and symbolic processing.


A tentative Schedule
---------------------

Day 1:

A. Introduction:

* Invited talks  
  These talks will provide an overview of the field and set the tone for 
  ensuing discussions. 
* Theoretical foundations for integrating connectionist and symbolic 
  processing

B. Definition of the two approaches: 

* Do they exhaust the space of current research in connectionist-symbolic 
  integration, or is there room for additional categories?
* How do we compare the unified and hybrid approaches?
* Do the unified and hybrid approaches constitute a clearcut dichotomy or
  are they just endpoints of a continuum?
* What class of processes and problems is well-suited to unified or hybrid 
  integration? The relevant motivations and  objectives.
* What type of model is suitable for what type of application? Enumerate 
  viable target domains.

C. State of the art:

* Recent or ongoing theoretical or experimental research work 
* Implemented models belonging to either the unified or hybrid approach
* Practical applications of both types of systems

Research addressing key issues concerning:

* the unified approach: theoretical or practical issues involving 
  systematicity, compositionality and variable binding, biologically 
  inspired models, connectionist knowledge representation, other 
  high-level connectionist models.

* the hybrid approach: modes and methods of coupling, task sharing between 
  various components of a hybrid system, knowledge representation and sharing.

* both: commonsense reasoning, natural language processing, analogical 
  reasoning, and more generally applications of unified and hybrid models.


Day 2:

D. Cognitive Aspects:

* Cognitive plausibility and relations to other AI paradigms
* In cognitive modeling, why should we integrate connectionist and symbolic 
  processing?
* Is there a clear cognitive rationale for such integration?  (we may
  need to examine in detail some typical areas, such as commonsense reasoning,
  and natural language processing)
* Is there psychological and/or biological evidence for existing models? 
  If so, what is it?

E. Open research issues:

* Can we now propose a common terminology with precise definitions for both 
  approaches to connectionist-symbolic integration and for the location on 
  the continuum?
* How far can unified systems go? Can unified models be supplemented by 
  hybrid models? Can hybrid models be supplanted by unified models?
* Limitations and barriers faced by both approaches
* What breakthroughs are needed for both approaches?
* Is it possible to synthesize various existing models? 


Workshop format
---------------
- panel discussions 
- mini-group discussions: participants will break into groups of 7/8
  to discuss a given theme; group leaders will then form a panel to
  report on group discussions and attempt a synthesis with audience
  participation 
- interactive talks: this is a novel type of oral presentation we will 
  experiment with. Instead of a classical presentation, the
  speaker will present a problem or issue and give a brief statement
  of his personal stand (5 min) to launch discussions which he will
  then moderate and conclude.
- classical slide talks followed by Q/A and discussions.


Workshop Co-chairs: 
-------------------
Frederic Alexandre, Crin-Cnrs/Inria-Lorraine
Ron Sun,   The University of Alabama 

Organizing Committee:
---------------------
John Barnden, New Mexico State University 
Steve Gallant,  Belmont Research Inc.       
Larry Medsker,  American University
Christian Pellegrini, University of Geneva
Noel Sharkey,  Sheffield University

Program Committee:
------------------
Lawrence Bookman (Sun Laboratory, USA)
Michael Dyer (UCLA, USA)
Wolfgang Ertel (FRW, Germany)
LiMin Fu (University of Florida, USA)
Jose Gonzalez-Cristobal (UPM, Spain)
Ruben Gonzalez-Rubio (University of Sherbrooke, Canada)
Jean-Paul Haton (Crin-Inria, France)
Melanie Hilario (University of Geneva, Switzerland)
Abderrahim Labbi (IMAG, France)
Ronald Yager (Iona College, USA)


Schedule:
---------
- The submission deadline for participants is February 1, 1995.
- The authors and potential participants will be notified the acceptance 
  decision by March 15, 1995.
- The camera-ready copies of working notes papers will be due on April 15, 1995


Submission:
-----------
- If you wish to present a talk, specify the preferred type of presentation 
  (classical or interactive talk) and submit 5 copies of an
  extended abstract (within the limit of 5-7 pages) to:

Prof. Ron Sun   
Department of Computer Science
The University of Alabama 
Tuscaloosa, AL 35487
(205) 348-6363

- If you only wish to attend the workshop, send 5 copies of a short
  (no more than one page) description of your interest to the same address 
  above.

- Please be sure to include your e-mail address in all submissions.



From marks at u.washington.edu  Thu Dec 22 00:55:40 1994
From: marks at u.washington.edu (Robert Marks)
Date: Wed, 21 Dec 94 21:55:40 -0800
Subject: Neural Networks in Russia
Message-ID: <9412220555.AA24404@carson.u.washington.edu>


For Connectionists:

              The 2-nd International Symposium
           on Neuroinformatics and Neurocomputers
                   Rostov-on-Don, RUSSIA
                   September 20-23, 1995

     Organized by Russian Neural Network Society (RNNS)

                          and

 A.B. Kogan Research Institute for Neurocybernetics (KRINC)

                   in co-operation with

     Institute of Electrical and Electronics Enginieers
             Neural Networks Council (IEEE NNC)


                   First Call for Papers

Research in Neuroinformatics and Neurocomputing continued in
Russia after  the research  was deflated  in the west in the
1970's.   The research sophistication in neural networks, as
a   result,  is   quite   advanced   in   Russia.  The first
international  RNNS/IEEE  Symposium, held  in October  1992,
proved  to  be  a  highly  successful  forum  for  a diverse
international  interchange  of   fresh  and  novel  research
results.    The    second   International    Symposium    on
Neuroinformatics   and   Neurocomputers   is   built on this
remarkable success.

The symposium  focus is  on the  neuroscience,  mathematics,
physics,  engineering  and  design  of  neuroinformatic  and
neurocomputing systems.

Rostov-on-Don, the  location of the Symposium, is about 1000
km south  of Moscow  on the  scenic Don  river.   The Don is
commonly identified  as the  boundary between the continents
of Europe  and Asia.   Rostov  is the home of the A.B. Kogan
Research Institute  for  Neurocybernetics  at  Rostov  State
University -  one of  the premier  neural  network  research
centers in Russia.

Papers for  the Symposium  should be  sent  in  CAMERA-READY
FORM, NOT  EXEEDING 8  PAGES in A4  format  to   the Program
Committee Co-Chair Alexander A. Frolov.   Two copies  of the
paper should  be submitted.

The deadline for submission is 15 MARCH, 1995.

Notification of acceptance will be sent on or before 15 May,
1995.




                    SYMPOSIUM COMMITTEE

GENERAL CHAIR

     Witali L. Dunin-Barkowski, Dr. Sci.,
     The 2-nd International Symposium
     on Neuroinformatics and Neurocomputers,
     Symposium Chair
     President of the Russian Neural Network Society,
     A.B. Kogan Research Institute for Neurocybernetics
     Rostov State University
     194/1 Stachka avenue,
     344104, Rostov-on-Don, Russia
     Tel: +7-8632-28-0588,
     Fax: +7-8632-28-0367
     E-mail: wldb at krinc.rostov-na-donu.su


PROGRAM COMMITTEE CO-CHAIRS

     Professor Alexander A. Frolov,
     The 2-nd International Symposium on Neuroinformatics and
        Neurocomputers,
     Program Co-Chair
     5a Butlerov str.
     Higher Nervous Activity and Neurophysiology Institute
     Russian Academy of Science
     117220, Moscow, RUSSIA.

     Professor Robert J. Marks II
     Program Co-Chair
     The 2-nd International Symposium on Neuroinformatics and
        Neurocomputers
     University of Washington
     Department of Electrical Engineering
     c/o 1131 199th Street S.W., Suite N
     Lynnwood, WA 98036-7138
     WA, USA.

Other information is available from the Symposium Committee.





From david at cns.ed.ac.uk  Thu Dec 22 03:58:07 1994
From: david at cns.ed.ac.uk (David Willshaw)
Date: Thu, 22 Dec 1994 08:58:07 GMT
Subject: elitism at NIPS
Message-ID: <199412220858.IAA09676@dumbo.cns.ed.ac.uk>


1. Even though the practice of removing the authors' names from
submitted papers before sending to referees is not perfect, in my view
it certainly helps to make the refereeing process less biased.

2. I would also support the suggestion that the cohort of referees be
broadened. One way of doing this is to make it more international. In
the list of 174 referees given in this year's programme of Abstracts,
I estimated that 84% (146) were from the USA and Canada. Only 6
referees were from Germany, 4 from France and 3 from the UK, for
example.

David



From trevor at media-lab.media.mit.edu  Thu Dec 22 11:01:55 1994
From: trevor at media-lab.media.mit.edu (Trevor Darrell)
Date: Thu, 22 Dec 94 11:01:55 EST
Subject: "Blind" reviews are impossible
In-Reply-To: ted@spencer.ctan.yale.edu's message of Mon, 19 Dec 1994 09:52:27 -0500 <199412191452.AA23787@PLANCK.CTAN.YALE.EDU>
Message-ID: <9412221601.AA06532@marblearch.media.mit.edu>

   Delivery-Date: Thu, 22 Dec 94 10:24:24 -0500
   Date: Mon, 19 Dec 1994 09:52:27 -0500
   From: ted at spencer.ctan.yale.edu

   Even a nonexpert reviewer can figure out who wrote a paper simply by
   looking for citations of prior work.  The only way to guarantee a
   "blind" review is to forbid authors from citing anything they've done
   before, or insist on silly euphemisms when citing such publications.

In computer vision papers have been reviewed blind at the major
conferences for the past several years. Authors are encouraged to
reference themselves in the third person. There seem to be
few complaints about the system. 

While it is sometimes possible to discern that a paper comes from a
particular school or group, it is usually impossible to know who was
the first author. And one can never be sure that the paper is not from
some other (unknown) person who has written the paper building
directly on the tradition of another group, and thus uses their
terminology but does not actually work there. In any case, can there
really be a disadvantage to blind reviewing? (It seems a weak point
indeed to say the correctness of papers in NIPS is assurred by their
authorship!)

Papers are not blind at the program committee level, so any agregious
mistakes brought on by blind reviewing (?) can be still corrected there. 

Just my $0.02 from the CV perspective...
--trevor


From trevor at mallet.Stanford.EDU  Thu Dec 22 18:10:03 1994
From: trevor at mallet.Stanford.EDU (Trevor Hastie)
Date: Thu, 22 Dec 1994 15:10:03 -0800
Subject: Report Announcement
Message-ID: <199412222310.PAA09264@mallet.Stanford.EDU>

The following report is available via anonymous ftp or Mosaic:

ftp://playfair.stanford.edu/pub/reports/hastie/dann.ps.Z

	Discriminant Adaptive Nearest Neighbor Classification

		Trevor Hastie and Robert Tibshirani

We propose an adaptive nearest neighbor rule that uses local
discriminant information to estimate an effective metric for
classification.  We also propose a method for global dimension
reduction, that combines local dimension information.  In a number of
examples, the methods demonstrate the potential for substantial
improvements over nearest neighbor classification.



From mozer at neuron.cs.colorado.edu  Thu Dec 22 18:32:47 1994
From: mozer at neuron.cs.colorado.edu (Michael C. Mozer)
Date: Thu, 22 Dec 1994 16:32:47 -0700
Subject: NIPS, blind reviewing, and elitism
Message-ID: <199412222332.QAA08554@neuron.cs.colorado.edu>

As NIPS*95 program chair, I want to respond to the issue of blind reviewing.
We have considered this idea over the past few months, and in balance the 
costs seem to outweigh the benefits.  Most of the arguments for and against 
blind reviewing were stated clearly in earlier messages.

NIPS is an elite conference in that researchers tend to self-select their best
work for submission, and even then only 25-30% of the submissions are accepted.
Further, researchers who do good work one year and have papers accepted are 
likely to do good work in the future, so it is not surprising that there is 
a core group of consistent contributors to the conference, even without any 
bias.  (Serving on the program committee before, I witnessed an opposite 
bias -- a bias against accepting multiple papers by an individual who 
had several strong submissions, and against awarding a talk to the same 
individual in successive years.)

Prior to 1994, there was likely some validity to the perception that NIPS 
"insiders" had an edge.  As a reviewer, it was difficult to evaluate work
solely on the basis of extended abstracts; in borderline cases, it helped to 
know the authors' track records.  After Dave Touretzky changed the submission
format to full papers in 1994, reviewers and program committee members to
whom I've spoken seemed satisfied that the reviewing process was objective.  
I suspect that the perception of unfairness may linger for a few years, but 
any such reality was squelched by the new submission format.

If you felt the reviewing process was not objective in 1994, I would like to 
hear your story.  (It is possible to communicate anonymously; to find out more 
about this service, send mail to help at anon.penet.fi.)  Several points of note 
before expressing a complaint:  (1) Many good submissions were ultimately 
rejected, simply because the submission pool tends to be of high quality and 
the number of accepted papers is limited by a maximum page count of the 
proceedings volume.  (2) Most of the people who have served as program and 
general chairs at NIPS have had papers rejected, including the current chairs!
The reviewers are the primary decision makers.  (3) Reviewers are not always
as competent and dilligent as one might like, although NIPS uses three 
reviewers per paper--plus an area chair to arbitrate--to mitigate the 
consequences of an inappropriate review.

I gladly welcome comments and suggestions aimed at broadening the constituency
of NIPS without lowering the meeting's quality.  I will summarize the feedback
I receive to the net.

Cheers, 

Mike Mozer


From postma at cs.rulimburg.nl  Fri Dec 23 05:26:23 1994
From: postma at cs.rulimburg.nl (Eric Postma)
Date: Fri, 23 Dec 94 11:26:23 +0100
Subject: Paper Announcement: Priming and Memory
Message-ID: <9412231026.AA28600@bommel.cs.rulimburg.nl>



------------------------------------------------------------------------
FTP-host: ftp.cs.rulimburg.nl
FTP-file: pub/papers/postma/memory.ps.Z
------------------------------------------------------------------------

The following paper is now available:


    The Nature of Memory Representations [6 pages]

Eric O. Postma, Ernst H. Wolf, H. Jaap van den Herik  
                        and
                Patrick T. W. Hudson

   Department of Computer Science, University of Maastricht 
      P.O.Box 616, 6200 MD Maastricht, The Netherlands 

To appear in the Proceeding of the workshop on Supercomputing in Brain
Research: From Tomography to Neural Networks, HLRZ, KFA Juelich,
Germany, November 21-23, 1994. World Scientific Publishing Company.

Abstract:
This study investigates processing and storage in the brain using a
priming task. We model a priming task with a relaxation neural network
- the Coulomb Energy Network. The network's performance on a
fragment-completion task is assessed by storing a set of memories into
the network and, subsequently, testing its completion performance.  It
has been claimed that the findings of stochastic independence on
repeated fragment completion imply noninteracting memory traces. We
nevertheless model memory in a way in which all traces embodying a
single representation interact and find stochastic independence when
the memories in the network are densely packed.  Dependence between
priming fragments can be obtained, but only when the memories are
sparsely packed.  We conclude that independence on repeated fragment
completion does not necessarily imply that the underlying memory traces
are noninteracting memory traces.

------------------------------------------------------------------------
Please do not reply directly to this message 
------------------------------------------------------------------------

FTP-instructions:

unix> ftp ftp.cs.rulimburg.nl
ftp> Name: anonymous
ftp> Password: your e-mail address
ftp> pub/papers/postma
ftp> binary
ftp> get memory.ps.Z
ftp> quit
unix> uncompress memory.ps.Z
------------------------------------------------------------------------

Merry Christmas,

Eric Postma
Computer Science Department
Faculty of General Sciences
University of Limburg
PO Box 616
6200 MD  Maastricht
The Netherlands

email: postma at cs.rulimburg.nl
web  : http://www.cs.rulimburg.nl
tel  : +31 43 883493
fax  : +31 43 252392


From jon at maths.flinders.edu.au  Fri Dec 23 06:58:43 1994
From: jon at maths.flinders.edu.au (Jonathan Baxter)
Date: Fri, 23 Dec 1994 22:28:43 +1030
Subject: TR Available: Learning Internal Representations
Message-ID: <199412231158.AA05637@calvin.maths.flinders.edu.au>

FTP-host: archive.cis.ohio-state.edu
FTP-filename: /pub/neuroprose/Thesis/baxter.thesis.ps.Z

--------------------------------------------------------

The following paper is now available:

Learning Internal Representations  [112 pages]
Jonathan Baxter

This is a preliminary draft of my PhD thesis. Note that it is in the
Thesis subdirectory of the neuroprose archive. It is in the process 
of being broken into several pieces for submission to Information and
Computation, Machine Learning and next year's COLT.

I'm afraid I cannot offer hard-copies.

ABSTRACT:
Most machine learning theory and practice is concerned with learning a
single task. In this thesis it is argued that in general there is insufficient
information in a single task for a learner to generalise well and that what
is required for good generalisation is information about {\em many
similar learning tasks}. The information about similar learning tasks forms a
body of prior information that can be used to constrain the hypothesis space
of the learner and cause it to generalise better.
Typical learning scenarios in which there are many similar tasks are image
recognition and speech recognition.

After proving that learning without prior information is impossible except in
the simplest of situations, the concept of the {\em environment} of a learner 
is introduced as a probability measure over the set of learning problems the
learner might be expected to learn. It is shown how a sample from such an
environment can be used to learn a {\em representation}, or recoding of 
the input space that is appropriate for the environment. Learning a
representation can equivalently be thought of as learning the
appropriate features of the environment.

Using Haussler's statistical decision theory framework for
machine learning, rigorous bounds are derived on the sample size
required to ensure good generalisation from a representation learning process.
These bounds show that under certain circumstances learning
a representation appropriate for $n$ tasks reduces the number of examples
required of each task by a factor of $n$. It is argued that environments
such as character recognition and speech recognition fall into the category
of learning problems for which such a reduction is possible.

Once a representation is learnt it can be used to learn {\em novel} tasks
from the same environment, with the result that far fewer examples are
required of the new tasks to ensure good generalisation. Rigorous bounds are
given on the number of tasks and the number of samples from each task required
to ensure that a representation will be a good one for learning novel tasks.

All the results on representation learning are generalised to cover any form
of automated hypothesis space bias that utilises information from similar
learning problems. 

It is shown how gradient-descent based
procedures for training Artificial Neural Networks can be generalised to cover
representation learning. Two experiments using the
new procedure are performed. Both experiments fully support the
theoretical results.

The concept of the environment of a learning process
is applied to the problem of {\em vector quantization} with the result that
a {\em canonical} distortion measure for the quantization process emerges.
This distortion measure is proved to be optimal if the task is to
approximate the functions in the environment.

Finally, the results on vector quantization are reapplied to
representation learning to yield an improved error measure for learning in
classifier environments. An experiment is presented demonstrating the
improvement. 

-------------

Retrieval Intructions:

unix> ftp archive.cis.ohio-state.edu
ftp> Login: anonymous
ftp> Password: e-mail address
ftp> cd pub/neuroprose/Thesis
ftp> binary
ftp> get baxter.thesis.ps.Z
ftp> quit
unix> uncompress baxter.thesis.ps.Z
unix> lpr -s baxter.thesis.ps
---------------

Jonathan Baxter
School of Information Science and Technology,
The Flinders University of South Australia.
jon at maths.flinders.edu.au



From 100020.2727 at compuserve.com  Fri Dec 23 11:24:19 1994
From: 100020.2727 at compuserve.com (Andrew Wuensche)
Date: 23 Dec 94 11:24:19 EST
Subject: Discrete Dynamics Lab (for DOS)
Message-ID: <941223162419_100020.2727_BHL46-1@CompuServe.COM>

Discrete Dynamics Lab 
---------------------
Announcing the first release of Discrete Dynamics Lab (described below),
a program for studying discrete dynamical networks, from Cellular
Automata to random Boolean networks, including their attractor basins. 
   
   Attractor basins are objects in space-time that link network states
according to their transitions. Access to these objects provide insights
into complexity, chaos and emergent phenomena in CA. In less ordered
networks (as well as CA), attractor basins show how the network
categorises its state space far from equilibrium, and represent the
network's memory.

   The program is released as shareware. This is a beta version and
comments are welcome. Section 1-10 in the ddlab.txt file gives an
overview of the program. Appendix A gives some operating instructions
including quick-start examples, but a detailed reference is not yet
complete. An illustrated operating manual will be released in due course.
For further background on attractor basins of CA and random Boolean
networks, and their implications, refer to refs. [1,2,3,4] below. [1] is
a book, hard copy pre-prints of [1,2,3] are available on request. 

  Platform - PC-DOS 386 or higher, with VGA or SVGA graphics, maths
co-processor, mouse, extended memory recomended. Ideally a fast 486 with
8MB ram.

download instructions
---------------------
ftp the file "ddlab.zip" (416,884 bytes):

%          ftp ftp.cogs.susx.ac.uk
name:      anonymous
password:  your complete e-mail address 
ftp>       cd pub/alife/ddlab
ftp>       binary
ftp        get ddlab.zip
ftp        close

unzip "ddlab.zip" to give the following files:

ddlab.txt    a text file describing ddlab
ddlab.exe    the program
dos4dw.exe   the DOS extender, giving access to extended memory
gliders.r_s  file with "glider" rules 
smalle.fon   two font files
sserife.fon    

to run, from the directory containing these files, enter: ddlab

All questions and comments to 

Andy Wuensche 

contact address:                      Santa Fe Institute
48 Esmond Road, London W4 1JQ UK      and The University of Sussex (COGS)
tel 081 995 8893  fax 081 742 2178    wuensch at santafe.edu
100020.2727 at compuserve.com            andywu at cogs.susx.ac.uk

Discrete Dynamics Lab
---------------------
Cellular Automata - Random Boolean Networks.
(copyright (c) Andrew Wuensche 1993)
First release of the beta version Dec 1994. 

DDLab is an interactive graphics program for research into the dynamics
of finite binary networks (for a DOS-PC platform). The program is
relevant to the study of complexity, emergent phenomena and neural
networks, and implements the investigations presented in [1,2,3,4].
   
   Using a flexible user interface, a network can be set up for any
architecture between regular CA (1d or 2d with periodic boundary
conditions) on the one hand[1], and random Boolean networks (disordered
CA) on the other[2]. The latter have arbitrary connections, and rules
which may be different at each site. The neighbourhood (or
pseudo-neighbourhood) size may be set from 1 to 9, and the network may
have a mix of neighbourhood sizes.
   
   The program iterates the network forward to display space-time
patterns (mutations are possible "on the fly"), and also runs the network
"backwards" to generate a pattern's predecessors and reconstruct its
branching sub-tree of all ancestor patterns until all "garden of Eden"
states, the leaves of the sub-tree, have been reached. For smaller
networks, sub-trees, basins of attraction or the whole basin of
attraction field can be displayed as a directed graph or set of graphs in
real time, with many presentation options. Attractor basins may be
"sculpted" towards a desired configuration. Various statistical and
analytical measures and numerical data are made available, mostly
displayed graphically. 

   The network's parameters, and the graphics display and presentation
options, can be very flexibly set, reviewed and altered. Changes can be
made "on the fly", including mutations to rules, connections or current
state. 2d networks (including the "game of life" or any mutation thereof)
can be displayed as a space-time pattern in a 3d isometric projection.
Network parameters, states, data, and the screen image can be saved and
loaded in a variety of tailor-made file formats. 

   Statistical measures and data (mostly presented graphically) include:
lambda and Z parameters, rule-table lookup frequency and entropy, pattern
density, detailed data on sub-trees, basins and fields, garden-of-Eden
density, in-degree frequency and a scatter plot of state-space.  
   
   Learning/forgetting algorithms allow attaching/detaching sets of
states as predecessors of a given state by automatically mutating rules
or changing connections. This allows "sculpting" the basin of attraction
field to approach a desired scheme of hierarchical categorisation. 
    
References. 
-----------
Wuensche,A., and M.J.Lesser. "The Global Dynamics of Cellular Automata:
An Atlas of Basin of Attraction Fields of One-Dimensional Cellular
Automata", Santa Fe Institute Studies in the Sciences of Complexity,
Reference Vol.I, Addison-Wesley, 1992.

Wuensche.A.,"The Ghost in the Machine: Basins of Attraction of Random
Boolean Networks", in Artificial Life III, Santa Fe Institute Studies in
the Sciences of Complexity, Addison-Wesley, 1994. 
 
Wuensche.A., "Complexity in One-D Cellular Automata: Gliders, Basins of
Attraction and the Z parameter", Santa Fe Institute working paper
94-04-025, 1994.

Wuensche.A., "The Emergence of Memory; Categorisation Far From
Equilibrium", Cognitive Science Research Paper 346, University of Sussex,
1994. To appear in "Towards a Scientific Basis for Consciousness" eds SR
Hameroff, AW Kaszniak, AC Scot, MIT Press. 




From G.Schram at ET.TUDelft.NL  Fri Dec 23 11:54:58 1994
From: G.Schram at ET.TUDelft.NL (Gerard Schram)
Date: Fri, 23 Dec 1994 17:54:58 +0100
Subject: Report Announcement
Message-ID: <01HKZPN1751S004ZXG@TUDERA.ET.TUDELFT.NL>

The following report is available via anonymous ftp or Mosaic:

ftp://playfair.stanford.edu/pub/reports/hastie/dann.ps.Z

	Discriminant Adaptive Nearest Neighbor Classification

		Trevor Hastie and Robert Tibshirani

We propose an adaptive nearest neighbor rule that uses local
discriminant information to estimate an effective metric for
classification.  We also propose a method for global dimension
reduction, that combines local dimension information.  In a number of
examples, the methods demonstrate the potential for substantial
improvements over nearest neighbor classification.



---------------------*********---------------------------
Gerard Schram (g.schram at et.tudelft.nl)
Control laboratory, Department of Electrical Engineering,
Delft University of Technology
P.O.Box 5031, 2600 GA Delft, The Netherlands
Phone:   +31-15-785114
Telefax: +31-15-626738
---------------------*********---------------------------



From eric at research.NJ.NEC.com  Fri Dec 23 11:59:24 1994
From: eric at research.NJ.NEC.com (Eric B. Baum)
Date: Fri, 23 Dec 94 11:59:24 EST
Subject: NIPS, blind reviewing, and elitism
Message-ID: <9412231659.AA00986@yin>

IMO the NIPS '94 program was the strongest NIPS program I've
seen, and in fact the strongest program I've seen in any
conference of even moderate size, in the sense that I saw
no weak or uninteresting papers. (However I only visited
a fraction of the posters so I may have missed some.)

Any skeptics should wait and consult the Proceedings,
which IMO will provide a Prima Facie Case that the
refereeing procedure is not broken, and should not be
fixed (at least radically).

-------------------------------------
Eric Baum
NEC Research Institute, 4 Independence Way, Princeton NJ 08540
PHONE:(609) 951-2712, FAX:(609) 951-2482, Inet:eric at research.nj.nec.com




From schmidhu at informatik.tu-muenchen.de  Fri Dec 23 12:43:29 1994
From: schmidhu at informatik.tu-muenchen.de (Juergen Schmidhuber)
Date: Fri, 23 Dec 1994 18:43:29 +0100
Subject: No subject
Message-ID: <94Dec23.184341met.42261@papa.informatik.tu-muenchen.de>


              SEMILINEAR PREDICTABILITY MINIMIZATION 
	   PRODUCES ORIENTATION SENSITIVE EDGE DETECTORS

       Technical Report FKI-201-94 (7 pages, 0.95 Megabytes) 
             Juergen Schmidhuber       Bernhard Foltin
                     Fakultaet fuer Informatik 
                  Technische Universitaet Muenchen  
                      80290 Muenchen, Germany 

			 December 24, 1994
           
    Static real world images are processed by a computationally 
    simple  and  biologically plausible  version of  the recent 
    predictability  minimization   algorithm  for  unsupervised 
    redundancy  reduction.  Without a  teacher and  without any 
    significant pre-processing, the system automatically learns 
    to generate  orientation  sensitive  edge  detectors in the 
    first (semilinear) layer.

To obtain a copy, do:

	     unix>         ftp flop.informatik.tu-muenchen.de  
	                   (or ftp 131.159.8.35)
	     Name:         anonymous
             Password:     your email address 
	     ftp>          binary
	     ftp>          cd pub/fki
             ftp>          get fki-201-94.ps.gz  (0.222 MBytes)
	     ftp>          bye
	     unix>         gunzip fki-201-94.ps.gz
	     unix>         lpr fki-201-94.ps

Alternatively, check out
http://papa.informatik.tu-muenchen.de/mitarbeiter/schmidhu.html
If your net browser does not know gzip/gunzip (which works better
than compress/uncompress), I can mail you an uncompressed 
postscript version (as a last resort).

Merry Christmas!
Juergen Schmidhuber



From wahba at stat.wisc.edu  Fri Dec 23 22:36:19 1994
From: wahba at stat.wisc.edu (Grace Wahba)
Date: Fri, 23 Dec 94 21:36:19 -0600
Subject: SS-ANOVA for `soft' classification-anct
Message-ID: <9412240336.AA14678@hera.stat.wisc.edu>

The following report is available via anonymous ftp or Mosaic:

ftp://ftp.stat.wisc.edu/pub/wahba/exptl.ssanova.ps.gz

Smoothing Spline ANOVA for Exponential Families, With Application
to the Wisconsin Epidemiological Study of Diabetic Retinopathy

        Grace Wahba, Yuedong Wang, Chong Gu
        Ronald Klein MD, and Barbara Klein MD

Given attributes (which may be discrete or continuous)
and outcomes (Class 1 or Class 0) of a sample of instances,
we develop Smoothing Spline ANOVA methods to estimate the 
*probability* of membership in Class 1, given the attribute 
vector. These methods are suitable when outcomes 
as a function of attributes are not clear-cut, as, for 
example, occurs when estimating risk of some medical outcome, 
given various predictor variables and treatments. 
These methods are penalized log-likelihood methods and 
plots of the cross sections of the estimates are generally
fairly easy to interpret in context. 
We use the results to estimate the probability
of four-year progression of diabetic retinopathy, given the three
predictor variables glycosylated hemoglobin, duration of diabetes and 
body mass index at the study baseline, based on data 
from the Wisconsin Epidemiological Study of Diabetic Retinopathy.
We discuss methods for multiple smoothing parameter selection, 
(the bias-variance tradeoff!!), numerical methods for 
computing the estimate and Bayesian `confidence intervals' 
for the estimate. We discuss methods of informal and formal
model selection (stacked generalization!!) and some open questions.
This work provides further details for work which has previously been 
announced in NIPS-93 and elsewhere.

Other related papers in the same directory include 
gacv.ps.gz, ssanova.ps.gz ml-bib.ps and theses/ywang.thesis.README
 
 Grace Wahba wahba at stat.wisc.edu .....`think snow'


From lazzaro at CS.Berkeley.EDU  Wed Dec 28 18:11:33 1994
From: lazzaro at CS.Berkeley.EDU (John Lazzaro)
Date: Wed, 28 Dec 1994 15:11:33 -0800
Subject: No subject
Message-ID: <199412282311.PAA15382@snap.CS.Berkeley.EDU>



	I sent these comments to Mike Mozer in response to his call
for NIPS comments here last week, and he suggested I pass my comments
on to the mailing list.

	As a EE, I tend to look towards IEEE conferences as examples:
there seems to be two major types.

[1] Conferences whose prime goal is to bring together as large a segment of
a research community as possible, while still maintaining a standard of
quality for presentations. IEEE Circuits and Systems and IEEE Acoustics,
Speech, and Signal Processing are two prime examples. Among other 
attributes, its expected that very many if not the majority of the 
attendees will be presenting, and that the conference's primary clientele is
the research community (both academic and industrial), as opposed to the
industrial developer.

[2] Conferences whose prime goal is to showcase the state of the art in
a field, primarily for the benefit of applied development attendees. In 
these conference, a large majority of the attendees will not be presenting
or even doing research, but are either applied developers or non-technical
observers. IEEE Solid State Circuits conference is probably the best 
example of this type of conference: it has a session on the N best new 
microprocessor designs, the N best new DRAM designs,ect. For some types of
chips, academic groups can be competitive with industry groups, and the
session is mixed; in others, a $100 million dollar investment is needed to
design the chip, and so industrial groups dominate.

        Many [1] conferences have explicit rules in submission to ensure as
many different research groups around the world are presenting
as possible: some limit the number of papers a single author can submit,
others use "membership" techniques to try to limit the number of papers from
a lab. On the other hand, [2] conferences are concerned with fairness and with
having the highest quality, but explicitly do not have "broadness" in their
charter: if the N best new Op-Amps all come from a certain company, so be it.

        NIPS has always seemed squarely in the middle of these two types
of conferences, with the "elite" aspirations of the Solid State Circuits
conference, but with a ratio of "researcher" to "developer" attendees that
is closer to the first type of conference. Certain other conferences (most
notably SIGGRAPH) started out where NIPS has been, and ended up as a [2]
type of conference -- even if you discount the majority of the SIGGRAPH
attendees who are there for the trade show and arty stuff, most attendees
of the conference papers are there to listen and learn new research ideas
for possible use in development, not to present papers themselves.      

        The tone of most of the postings on this thread seem to imply that
[1] is the direction that they'd like NIPS to move to. If so, the IEEE
experience has been that more direct approaches than "blind reviewing" can
be used to broaden the conference.

        Personally, I'd prefer [2], because I believe the technologies 
associated with NIPS are going to have the same degree of engineering 
impact as SIGGRAPH (Computer Graphics) and ISSCC (Integrated Circuits) has
had on the world, and part of realizing that impact is having a conference
of type [2]. But for this to occur, NIPS needs to market itself to the
product development community, so that the ratio of developer:researcher
at NIPS increases significantly.





From ronnyk at CS.Stanford.EDU  Sat Dec  3 10:26:20 1994
From: ronnyk at CS.Stanford.EDU (Ronny Kohavi)
Date: Sat, 3 Dec 1994 23:26:20 +0800
Subject: MLC++: Machine Learning Utilities Available
Message-ID: <9412040726.AA14178@starry.Stanford.EDU>


			   MLC++ Utilities                
                           _______________


MLC++ is a Machine Learning library of C++ classes being developed at
Stanford.  More information about the library can be obtained at URL
http://robotics.stanford.edu:/users/ronnyk/mlc.html.

We are now releasing the object code for some utilities written using MLC++.
These will run on Suns, either on SunOS or Solaris.  


Included in the current release are the following induction algorithms:

1. Majority (baseline).
2. Basic ID3 for inducing decision trees.  The output can be sent
     to a mail server to get a postscript picture of the resulting
     tree.  Very useful for looking at the final tree and for teaching.
3. Basic nearest neighbor.
4. Decision Table.
5. Interface to C4.5 (for utilities below).
6. Feature subset selection : wraps around any of the above and
     selects a good subset of the features, usually improving
     performance and comprehensibility.
  


Utilities released are:

1. Cross validation : cross validate a file and any of the above induction
   algorithms.  Allows regular or stratified CV.
   You can also generate the cross validation files to compare
   your own induction algorithm.
  
2. Learning curve : generate a learning curve for any of the above
   induction algorithms.

3. Project : project the data onto a subset of attributes.

4. Convert : convert nominal attributes to unary encoding or binary
   encoding.


Quick starter guide:
--------------------

The MLC++ utilities are accessible by anonymous ftp to
  starry.stanford.edu:pub/ronnyk/mlc/
There are currently two kits, one for Solaris (MLCutil-solaris.tar.Z)
   and one for SunOS (MLCutil-sunos.tar.Z).

cd <directory>
zcat <kit-name> | tar xvf -

where <directory> is the directory under which the mlc directory will
   be built (e.g., /usr/local), and <kit-name> is the kit appropriate
   for your machine.

The documentation is in utils.ps.
The environment variable MLCDIR must be set to the directory where
    the utilities are installed.

Databases in the MLC++ format, which is very similar to C4.5 format
  can be found in starry.stanford.edu:pub/ronnyk/mlc/db.
  Most datafiles are converted from the repository at UC Irvine.

Questions or help requests related to the utilities should be
    addressed to mlcpp-help at CS.Stanford.EDU

--

   Ronny Kohavi (ronnyk at CS.Stanford.EDU, http://robotics.stanford.edu/~ronnyk)



From marcus at hydra.ifh.de  Mon Dec  5 04:37:52 1994
From: marcus at hydra.ifh.de (Marcus Speh)
Date: Mon, 5 Dec 94 10:37:52 +0100
Subject: Paper - GAUGE THEORY OF THINGS ALIVE AND UNIVERSAL DYNAMICS
Message-ID: <9412050937.AA21444@hydra.ifh.de>

The following paper is now available via anonymous FTP (see below):

GAUGE THEORY OF THINGS ALIVE AND UNIVERSAL DYNAMICS
BY G. MACK, Theoretical Physics, University of Hamburg, Germany

SOURCE FORMAT: 13 pages, latex, uses fleqn.sty (can be removed without harm)
REPORT-NO: DESY 94-184

				ABSTRACT

  Positing complex adaptive systems made of agents with relations
  between them that can be composed, it follows that they can be
  described by gauge theories similar to  elementary
  particle theory and  general relativity. By definition, a universal
  dynamics is able to determine the time development of any such system
  without need for further specification. The possibilities are limited,
  but one of them - reproduction fork dynamics - describes DNA
  replication and is the basis of biological life on earth. It is a
  universal copy machine and a renormalization group fixed point.
  A universal equation of motion in continuous time is also presented.


TO RETRIEVE, connect as USER "anonymous" with PASSWORD "<userid>@<address>" 

a) in Europe, to 
   ftp.desy.de (131.169.30.42) and retrieve one of the files from
   directory "pub/outgoing"

 203657 Dec  5 10:06 DESY_94-184.ps     [PostScript, uncompressed]
  60601 Dec  5 10:06 DESY_94-184.ps.gz  [PostScript, compressed]
  48393 Dec  1 14:48 DESY_94-184.tex    [LaTeX source]

OR in the US to 

b) ftp.scri.fsu.edu (144.174.128.34) and retrieve the LaTeX source
   file from directory "hep-lat/papers/9411":

  48506 Nov 28 07:27 9411059

On the World-Wide Web, you can also access the URL

  http://www.desy.de/pub/outgoing/DESY_94-184.ps.gz  [60KB]



Inquiries and comments should be sent directly to the author,
Prof. G. Mack <i02mac at dsyibm.desy.de>




From wuertz at ips.id.ethz.ch  Mon Dec  5 08:45:23 1994
From: wuertz at ips.id.ethz.ch (Diethelm Wuertz)
Date: Mon, 05 Dec 1994 14:45:23 +0100
Subject: PASE 95: Parallel Applications in Statistics and Economics
Message-ID: <9412051433.AA11623@sitter>


                            5th Anniversary
                          First Announcement
                            
                       International Workshop on 
           Parallel Applications in Statistics and Economics

                     >> Non-linear Data Analysis <<

                         Trier - Mainz, Germany
                      August 29 - September 2, 1995

____________________________________________________________________________
PURPOSE OF THE WORKSHOP:

  The purpose of this workshop is to bring together researchers interested 
  in innovative information processing systems and their applications in the 
  areas of statistics, finance and economics. The focus will be on in-depth 
  presentations of state-of-the-art methods and applications as well as on 
  communicating current research topics. This workshop is intended for 
  industrial and academic persons seeking new ways of comprehending the 
  behavior of complex dynamic systems. The PASE'95 workshop is concerned 
  with but not restricted to the following topics:

   o  Applications in finance, economics, and natural science
   o  Statistical tests for finding deterministic and chaotic behavior
   o  Statistical tests for measuring stability and stationarity
   o  Sampling and retrieving techniques for high frequency data
   o  Modeling and analysis of non-linear multivariate time series
   o  Statistical use of neural networks, genetic algorithms and fuzzy systems

____________________________________________________________________________
WORKSHOP SITE:

  The workshop will be held on a comfortable ship running from Trier to 
  Mainz on the scenic rivers Mosel and Rhine in Germany. The best connection 
  to Trier is to fly to Frankfurt and then to take the train. It will take 
  about 2 1/2 hours. From Mainz there are direct trains to Frankfurt Airport 
  (~ 30 minutes). Detailed traveling possibilities will be announced later.

____________________________________________________________________________
ACCOMMODATION AND REGISTRATION:
 
  All participants will be accommodated on the ship. The price of the 
  accommodation in a double room for 4 nights is Sfr 870.- for the main 
  deck (920.- for the upper deck) including breakfasts, lunches and dinners. 
  The welcome and workshop dinner are also included. The service begins 
  on Tuesday afternoon, August 29th, in Trier with registration and ends 
  on Saturday morning, September 2nd in Mainz. Since the facilities of 
  the ship are dedicated to the workshop the price is unique and no
  reductions can be provided for a shorter participation. Since the number 
  of participants is limited to about 100 persons, early registration is
  recommended. All applications will be handled on the basis "first come 
  first serve". Registration will only be accepted after a pre-payment of 
  Sfr 300.-. The rest of the accommodation expenses of Sfr 570.- or 620.-, 
  respectively, and the workshop fees must be transferred to our account

             PASE Workshop - Dr. Diethelm Wuertz
             Schweizerischer Bankverein, Zurich, Switzerland
             Number of Account: P0-206066.0

  before May 31st 1995. No currencies other than Swiss Francs will be 
  accepted.  In the case of a later remittance we cannot guarantee your
  application.

____________________________________________________________________________
WORKSHOP FEE AND PROCEEDINGS:

  The workshop fee is Sfr 580.- for profit-making companies and Sfr 230.- 
  for others. It includes the Proceedings, published as a Special Issue of 
  the Journal "Neural Network World - International Journal on Neural and 
  Mass-Parallel computing and Information Systems".

____________________________________________________________________________
WORKSHOP SCHOLARSHIPS:

  For students and a limited number of participants from "post-communist" 
  European countries some support and scholarships will be available: 
  please contact the organizers.

____________________________________________________________________________
ABSTRACTS AND DEADLINES:

  Regular abstracts of one page (approximately 30 lines with 60 characters) 
  must be submitted before December 1st, 1994. (For post-deadline papers
  please contact directly: <pase at ips.id.ethz.ch>.)  All contributions will be 
  refereed. The deadline for the full papers (about 10 pages) is April 1st, 
  1995. We cannot guarantee that papers submitted later will be printed in 
  the proceedings. The proceedings will be available at the conference. It 
  is also possible to submit an extended abstract of 4 pages. From these 
  contributions we will select, besides the regular invited speakers, 
  3 further speakers for an invited talk. The deadlines are the same as 
  in the case of regular abstracts. If you plan a soft- or hardware 
  demonstration please contact the organizers. Please send the abstracts 
  and full papers to:

    Hynek Beran, ICS Prag
    Pod vodarenskou vezi 2                FAX:    +42 2 858 57 89
    182 07 PRAGUE 8, Czech Republic       E-mail: pase at uivt.cas.cz

____________________________________________________________________________
ORGANIZATION:

  The Workshop will be organized by the Interdisciplinary Project Center 
  for Supercomputing (ETH Zurich), Olsen & Associates (Research Institute 
  for Applied Economics, Zurich) and the Institute of Computer Science 
  (Academy of Sciences, Prague)

    W.M. van den Bergh, D.E. Baestaens, Erasmus University Rotterdam
    Thilo von Czarnowski, Helaba Frankfurt
    Michel M. Dacorogna, Olsen & Associates Zurich
    Susanne Fromme, SMH Research Frankfurt
    Heinz Muehlenbein, GMD Sankt Augustin
    Gholamreza Nakkaeizadeh, Daimler Benz Forschung Ulm
    Paul Ormerod, Henley Center for Forecasting London
    Emil Pelikan, ICS Czech Academy of Sciences Prague
    Heinz Rehkugler, University of Bamberg
    Marco Tomassini, CSCS Manno
    Dieter Wenger, Swiss Bank Corporation Basel
    Diethelm Wuertz,  IPS ETH Zurich
    Hans Georg Zimmermann, Siemens AG, Munchen

____________________________________________________________________________
FURTHER INFORMATION:

  Further information will be available from anonymous ftp "maggia.ethz.ch"
  (129.132.17.1) or world wide webb "http://www.ips.id.ethz.ch/PASE/pase.html"

____________________________________________________________________________
REGISTRATION FORM - PASE '95:


   ....................................................................
   Mr./Mrs.        First Name        		Surname

   ....................................................................
   Institution or Company

   ....................................................................


   ....................................................................
   Mailing Address

   ....................................................................
   E-mail 

   ....................................................................
   Phone             			 Fax

   o  Fees University Sfr 230.--              o  Main Deck Sfr 870.-
   o  Fees Profit making Company Sfr 580.--   o  Upper Deck Sfr 920.-

   CONTRIBUTION:
   I would like to present a talk and to submit a paper with the 
   following title:

   ....................................................................

   ....................................................................

   ....................................................................

   Please, return this form as soon as possible to:

     Martin Hanf
     IPS CLU B2, ETH Zentrum                
     CH-8092 Zurich, Switzerland            
     E-Mail: pase at ips.id.ethz.ch

____________________________________________________________________________
PASE '95







--------------------------------------------------------------------------
PD Dr. Diethelm Wuertz
Interdisciplinary Project Center for Supercomputing
IPS CLU B3 - ETH Zentrum
CH-8092 ZURICH, Switzerland

Tel. +41-1-632.5567
Fax. +41-1-632.1104
--------------------------------------------------------------------------



From mbrown at aero.soton.ac.uk  Mon Dec  5 05:08:01 1994
From: mbrown at aero.soton.ac.uk (Martin Brown)
Date: Mon, 5 Dec 94 10:08:01 GMT
Subject: neural -probability-fuzzy seminars
Message-ID: <7370.9412051008@aero.soton.ac.uk>

                        ADT '95
                  Date: April 3-5 1995
        Venue: BRUNEL - The University of West London


UNICOM Seminars Ltd., Brunel Science Park,
Cleveland Road, Uxbridge, Middlesex, UB8 3PH, UK
Telephone: +44 1895 256484  FAX +44 1895 813095

ADT 95
Sponsored by:
Department of Trade and Industry (DTI)
Institute of Mathematics and its Applications
IFSA
International Neural Networks Society
European Neural Networks Society
AI Watch
Journal of Neural Computing and Applications
Rapid Data Ltd

Call for Papers, Registration Information, and Abstract Form

An international team of experts describes and explores the
most recent developments and research in Neural Networks,
Bayesian Belief Networks, Modern Heuristic Search Methods 
and Fuzzy Logic. Presentations cover theoretical developments 
as well as industrial and commercial applications.

Main Topics

Neural Networks and their Applications:
	New Developments in Theory and Algorithms; Simulations
	and Hardware; Pattern Recognition; Medical Diagnosis.

Modern Heuristic Search Methods:
	Genetic Algorithms; Classifier Systems; Simulated 
	Annealing; Tabu Search; Hybrid Techniques; Genetic and
	Evolutionary Programming.

Probabilistic Reasoning and Computational Learning:
	Bayesian Belief Networks; Computational Learning;
	Inductive Inference; Statistical Causality in Belief 
	Networks; Applications in Computational Vision, Medicine,
	Legal Reasoning, Fault Diagnosis and Genetics; Model 
	Selection.

Fuzzy Logic:
	Neural Fuzzy Programming; Algorithms and Applications; 
	Fuzzy Control; Fuzzy Data Processing; Fuzzy Silicon Chips.


PROGRAMME COMMITTEE & SPEAKERS:

Neural Networks and their Applications: 
	J G Taylor, King's College London (Chair); I Aleksander,
	Imperial College; M J Denham, University of Plymouth; 
	B Schuermann, Siemens AG; A G Carr, ERA Technology; 
	R Wiggins, DTI. Other Speakers include: S Hancock, Neural 
	Technologies Ltd; J Stonham, Brunel University, 
	J Austin, Univ. of York; B Clements, Clement Neuronics

Modern Heuristic Search Methods:
	V Rayward-Smith, University of East Anglia (Chair); 
	I Osman, University of Kent; C R Reeves, University of
	Coventry; G D Smith, University of East Anglia.
	Other Speakers include: E Aarts, Philips Research, 
	Netherlands; C Hughes, Logica; S Voss, Darmstadt Univ,
	D Schaaffer, Philips, USA; R Battiti, Trento Univ, Italy

Probabilistic Reasoning & Computational Learning:
	A Gammerman, Royal Holloway University of London (Chair);
	J Pearl, UCLA; D Spiegelhalter, Medical Research Council;
	A Dempster, Harvard University; C S Wallace, Monash
	University (Australia); V Vapnik, AT&T Bell Labs; 
	V Uspensky, University of Moscow.  Other Speakers include:
	P Reverberi, Inst. di Analisi dei Sistemi..Rome; M Talamo, Univ 
	degli Studi, Rome; S McClean, Ulster Univ; A Shen, Ind. Univ. 
	of Moscow; J Kwaan, Marconi Simulation

Fuzzy Logic:
	E H Mamdani, Queen Mary & Westfield College (Chair); J Baldwin,
	Bristol University; C Harris, University of Southampton;
	H Zimmerman, RWTH Aachen.Other Speakers include: L Zadeh, 
	Univ of Calif., Berkeley; R Weber, MIT GmbH; R John, De Montfort
	Univ.; H Bersini, IRIDIA, Belgium.


HOW TO CONTRIBUTE:

PRESENTATIONS:
You may wish to present a paper at any of these sessions. 
Please indicate your choice of session with your submission.
Please ensure we receive the title of your proposed talk and a short 
abstract (30-40 words) by 14 January 1995.  The Organising Committee 
will make their selections and we will advise you by 14 February 1995
whether you presentation can be included.  Written papers for the 
proceedings are also welcome. Please include your telephone and fax
numbers and postal address in all correspondence.

PUBLICATIONS:  Preliminary proceedings of this symposium and also
a refereed volume will be published.

Deadlines for Abstracts:  14 January 1995 - acceptance of paper
notified by 14 February 1995.  Extended abstracts (1000-1500) words,
by 15 March 1995.  Submission of papers for refereed proceedings
(3000-5000) words by 31 August 1995.

REGISTRATION: The conference program and registration material will be
available in late February 1995. To ensure receiving your copy,
complete the attached form and return it to ADT 95.

Registration Fee: GBP175 (Academic), GBP350 (Industrial), GBP500-800
(Exhibition), and GBP1000 Commercial Sponsors, preliminary proceedings
included in the fee.  
Late registration after 28 February 1995 will be charged GBP25 extra.


     -------------------------------------------------------------------
Please return the form below to:

ADT 95, UNICOM Seminars Ltd.,
Brunel Science Park, Cleveland Road, Uxbridge, Middlesex, UB8 3PH.
Telephone: +44 1895 256484.
FAX: +44 1895 813095.
Email: adt95 at unicom.demon.co.uk.

    --------------------------------------------------------------------

Name:
Organization:
Position / Department:

Address:




Telephone:
Facsimile:
E-Mail:

Please send me information about the event,

[ ] I wish to present a paper, title and abstract attached

[ ] I am interested in exhibiting

[ ] Send me more information about ADT95


This event is part of Brunel University's Continuing Education
Programme.


If you fail to return this form, you may not receive further
information.


From dayan at cs.TORONTO.EDU  Mon Dec  5 17:40:37 1994
From: dayan at cs.TORONTO.EDU (Peter Dayan)
Date: Mon, 5 Dec 1994 17:40:37 -0500
Subject: The Helmholtz Machine and the Wake-Sleep Algorithm
Message-ID: <94Dec5.174037edt.317@neuron.ai.toronto.edu>

FTP-host: ftp.cs.toronto.edu
FTP-filename: pub/dayan/wake-sleep.ps.Z
FTP-filename: pub/dayan/helmholtz.ps.Z

Two papers about stochastic and deterministic Helmholtz machines are
available in compressed postscript by anonymous ftp from
ftp.cs.toronto.edu - abstracts are given below.

We regret that hardcopies are not available. 

----------------------------------------------------------------------

pub/dayan/wake-sleep.ps.Z

      The Wake-Sleep Algorithm for Unsupervised Neural Networks
				   
  Geoffrey E Hinton, Peter Dayan, Brendan J Frey and Radford M Neal

Abstract:

We describe an unsupervised learning algorithm for a multilayer
network of stochastic neurons.  Bottom-up "recognition" connections
convert the input into representations in successive hidden layers and
top-down "generative" connections reconstruct the representation in
one layer from the representation in the layer above.  In the "wake"
phase, neurons are driven by recognition connections, and generative
connections are adapted to increase the probability that they would
reconstruct the correct activity vector in the layer below.  In the
"sleep" phase, neurons are driven by generative connections and
recognition connections are adapted to increase the probability that
they would produce the correct activity vector in the layer above.


Submitted for publication

----------------------------------------------------------------------

pub/dayan/helmholtz.ps.Z

			The Helmholtz Machine

  Peter Dayan, Geoffrey E Hinton, Radford M Neal and Richard S Zemel

Abstract:

Discovering the structure inherent in a set of patterns is a
fundamental aim of statistical inference or learning.  One fruitful
approach is to build a parameterised stochastic generative model,
independent draws from which are likely to produce the patterns.  For
all but the simplest generative models, each pattern can be generated
in exponentially many ways. It is thus intractable to adjust the
parameters to maximize the probability of the observed patterns, We
describe a way of finessing this combinatorial explosion by maximising
an easily computed lower bound on the probability of the observations.
Our method can be viewed as a form of hierarchical self-supervised
learning that may relate to the function of bottom-up and top-down
cortical processing pathways.


Neural Computation, in press.

----------------------------------------------------------------------


From lazzaro at CS.Berkeley.EDU  Mon Dec  5 17:44:49 1994
From: lazzaro at CS.Berkeley.EDU (John Lazzaro)
Date: Mon, 5 Dec 1994 14:44:49 -0800
Subject: No subject
Message-ID: <199412052244.OAA16160@boom.CS.Berkeley.EDU>


=========================== TR announcement =================================


REMAP: Recursive Estimation and Maximization of A Posteriori Probabilities
---Applications to Transition-based Connectionist Speech Recognition---

by H. Bourlard, Y. Konig & N. Morgan
   Intl. Computer Science Institute
   1947 Center Street, Suite 600
   Berkeley, CA 94704 
   email: bourlard,konig,morgan at icsi.berkeley.edu

ICSI Technical Report TR-94-064

Abstract

In this report, we describe the theoretical formulation of REMAP, an
approach for the training and estimation of posterior probabilities
using a recursive algorithm that is reminiscent of the EM (Expectation
Maximization) algorithm for the estimation of data likelihoods. Although
very general, the method is developed in the context of a statistical
model for transition-based speech recognition using Artificial Neural
Networks (ANN) to generate probabilities for hidden Markov models (HMMs).
In the new approach, we use local conditional posterior probabilities
of transitions to estimate global posterior probabili- ties of word
sequences given acoustic speech data. Although we still use ANNs to
estimate posterior probabilities, the network is trained with targets
that are themselves estimates of local posterior probabilities. These
targets are iteratively re-estimated by the REMAP equivalent of the
forward and backward recursions of the Baum-Welch algorithm to
guarantee regular increase (up to a local maximum) of the global
posterior probability. Convergence of the whole scheme is proven.

Unlike most previous hybrid HMM/ANN systems that we and others have
developed, the new formulation determines the most probable word
sequence, rather than the utterance corresponding to the most probable
state sequence. Also, in addition to using all possible state
sequences, the proposed training algorithm uses posterior probabilities
at both local and global levels and is discriminant in nature.


The postscript file of the full technical report (66 pages) can be copied
from our (anonymous) ftp site as follows:
ftp ftp.icsi.berkeley.edu
username= anonymous
passw= your email address
cd pub/techreports/1994
binary
get tr-94-064.ps.Z


		



From esann at dice.ucl.ac.be  Tue Dec  6 04:33:59 1994
From: esann at dice.ucl.ac.be (esann@dice.ucl.ac.be)
Date: Tue, 6 Dec 1994 11:33:59 +0200
Subject: Neural Processing Letters Vol.1 No.2
Message-ID: <9412061028.AA06065@ns1.dice.ucl.ac.be>

The following articles may be found in the second issue of the "Neural
Processing Letters" journal (November 1994):

- Good teaching inputs do not correspond to desired responses in ecological
  neural networks
     S. Nolfi, D. Parisi
- Recurrent Sigma-Pi-linked back-propagation network
     T.W.S. Chow, G. Fei
- An evolutionary approach to associative memory in recurrent neural
  networks
     S. Fujita, H. Nishimura
- Equivalence between some dynamical systems for optimization
     K. Urahama
- VISOR: schema-based scene analysis with structured neural networks
     W.K. Leow, R. Miikkulainen
- Nonlinear neural controller with neural Smith predictor
     Y. Tan, A. Van Cauwenberghe
- A neural method for geographical continuous field estimation
     D. Pariente, S. Servigne, R. Laurini

Neural Processing Letters in a rapid publication journal, aimed to publish
new ideas, original developments and work in progress in all aspects of the
field of Artificial Neural Networks.  The delay between submission of
papers and publication is maximum about 3 months.  Please don't hesitate to
ask for the instructions for authors, in order to submit your work to
Neural Processing Letters.

We remind you that all information concerning this journal may be found on
the following servers:
- FTP server: ftp.dice.ucl.ac.be
  directory: /pub/neural-nets/NPL
  login: anonymous
  password: your_e-mail_address
- WWW server: http://www.dice.ucl.ac.be/neural-nets/NPL/NPL.html

For any information (subscriptions, instructions for authors, free sample
copies...), you can also directly contact the publisher:
           D facto publications
           45 rue Masui
           B-1210 Brussels
           Belgium
           Phone: + 32 2 245 43 63
           Fax:   + 32 2 245 46 94






_____________________________
D facto publications -
        conference services
45 rue Masui
1210 Brussels
Belgium
tel: +32 2 245 43 63
fax: +32 2 245 46 94
_____________________________




From mario at physics.uottawa.ca  Tue Dec  6 09:18:06 1994
From: mario at physics.uottawa.ca (Mario Marchand)
Date: Tue, 6 Dec 94 10:18:06 AST
Subject: Neural net paper available by anonymous ftp
Message-ID: <9412061418.AA24196@physics.uottawa.ca>

The following paper, which was presented at the NIPS'94 conference,
is available by anonymous ftp at:

ftp://dirac.physics.uottawa.ca/usr2/ftp/pub/tr/marchand


FileName: nips94.ps

Title: Learning Stochastic Perceptrons Under k-Blocking Distributions

Authors: Marchand M. and Hadjifaradji S.

Abstract:
We present a statistical method
that PAC learns the class of stochastic
perceptrons with arbitrary monotonic activation function
and weights $w_i \in \{-1, 0, +1\}$
when the probability distribution that generates
the input examples is member of a family that we call {\em k-blocking
distributions\/}. Such distributions
represent an important step beyond the case
where each input variable is statistically independent
since the 2k-blocking family contains all the Markov
distributions of order k.
By stochastic perceptron we mean a perceptron which,
upon presentation of input
vector $\x$,  outputs~1 with probability  $f(\sum_i w_i x_i - \theta)$.
Because the same algorithm works for any monotonic
(nondecreasing or nonincreasing) activation function $f$
on Boolean domain,
it handles the well studied cases of sigmo\"{\i}ds
and the ``usual'' radial basis functions.

ALSO: you will find other papers co-authored by Mario Marchand in
      this directory. The text file: Abstracts-mm.txt contains a
      list of abstracts of all the papers.

PLEASE: communicate to me any printing or transmission problems.
        Any comments concerning these papers are very welcome.




   ----------------------------------------------------------------
  |  UUU            UUU            Mario Marchand                  |
  |  UUU            UUU            -----------------------------   |
  |  UUU       OOOOOOOOOOOOOOOO    Department of Physics           |
  |  UUU      OOO   UUU      OOO   University of Ottawa            |
  |   UUUUUUUUUUUUUUUU       OOO   150 Louis Pasteur street        |
  |           OOO            OOO   PO BOX 450 STN A                |
  |            OOOOOOOOOOOOOOOO    Ottawa (Ont) Canada K1N 6N5     |
  |                                                                |
  |   ***** Internet E-Mail: mario at physics.uottawa.ca **********   |
  |   ***** Tel: (613)564-9293 ------------- Fax: 564-6712 *****   |
   ----------------------------------------------------------------





From J.Vroomen at kub.nl  Tue Dec  6 12:58:06 1994
From: J.Vroomen at kub.nl (Jean Vroomen)
Date: 6 Dec 94 12:58:06 MET
Subject: PhD research position in Cogn Psy/Comput Linguistics at Tilburg
Message-ID: <MAILQUEUE-101.941206125806.384@fsw_1.kub.nl>

Predoctoral research associate position in
cognitive psychology/computational linguistics
(Ph.D position, 4 year contract)
Tilburg University

A predoctoral position in cognitive psychology/computational
linguistics is available immediately at Tilburg University. The
focus of research will be on the development of a text-to-speech
conversion system with special emphasis on modelling reading ac-
quisition and developmental dyslexia. The development of the model
and the simulations will be conducted together with behavioral
experiments.

In the project, a strong cross-linguistic approach is taken.
Comparative models will be built for English, French, and Dutch. This
approach furthermore includes the development of a computational
measure of the complexity of these writing systems (`orthographic
depth').

Candidates with training in cognitive psychology, cognitive neurop-
sychology combined with computational (connectionist) modelling are
preferred.

Applicants should send a curriculum vitea and a brief description
of fields of interest to:

Professor Beatrice de Gelder
Department of Psychology
Tilburg University
Warandelaan, 2
PO Box 90153
5000 LE Tilburg, The Netherlands
email: b.degelder at kub.nl


From FRYRL at f1groups.fsd.jhuapl.edu  Wed Dec  7 15:30:00 1994
From: FRYRL at f1groups.fsd.jhuapl.edu (Fry, Robert L.)
Date: Wed, 07 Dec 94 12:30:00 PST
Subject: New Neuroprose Entry
Message-ID: <2EE61B8E@fsdsmtpgw.fsd.jhuapl.edu>


     A preprint of a manuscript entitled "Observer-participant models of 
neural computation" which has been accepted by the IEEE Trans. Neural 
Networks is being made available via FTP from the Neuroprose directory. 
 Details for retrieving this article an associated figures follows the 
abstract below:

               ABSTRACT

     Observer-participant models of neural computation

          A model is proposed in which the neuron serves as an information 
channel. Channel distortion occurs through the channel since the mapping 
from input boolean codes to output codes are many-to-one in that neuron 
outputs consist of just two distinguished states. Within the described 
model, the neuron performs a decision-making function.  Decisions are made 
regarding the validity of a question passively posed by the neuron to its 
environment.  This question becomes defined through learning hence learning 
is viewed as the process of determining an appropriate question based on 
supplied input ensembles.  An application of the Shannon information 
measures of entropy and mutual information taken together in the context of 
the proposed model lead to the Hopfield neuron model with conditionalized 
Hebbian learning rules implemented through a simple modification to Oja's 
learning equation.  Neural decisions are shown to be based on a  sigmoidal 
transfer characteristic or in the limit as computational temperature tends 
to zero, a maximum likelihood decision rule.  The described work is 
contrasted with the information-theoretic approach of Linsker.


The paper is available in two files from archive.cis.ohio-state.edu in the
/pub/neuroprose  subdirectory.  The file names are

     fry.maxmut.ps.Z   (compressed postscript manuscript 433927 bytes)
     fry.maxmut_figs.ps.Z (comp. postscript figures 622321 bytes)

Robert L. Fry
Johns Hopkins University/
Applied Physics Laboratory
Johns Hopkins Road
Laurel, MD 20723
robert_fry at jhuapl.edu


From rohwerrj at helios.aston.ac.uk  Wed Dec  7 12:21:49 1994
From: rohwerrj at helios.aston.ac.uk (rohwerrj)
Date: Wed, 7 Dec 1994 17:21:49 +0000
Subject: 2 tech reports on n-tuple classifiers
Message-ID: <22434.9412071721@sun.aston.ac.uk>

FTP-host: archive.cis.ohio-state.edu
FTP-filename: /pub/neuroprose/morciniec.ntup_bench.ps.Z
FTP-filename: /pub/neuroprose/rohwer.ram_bayes.ps.Z

The following two tech reports have been placed on the Neuropose
ftp site:
     archive.cis.ohio-state.edu (128.146.8.52)
  in pub/neuroprose/

Copies are also available on the Aston University ftp site:
      cs.aston.ac.uk (134.151.52.106)
   in pub/docs/

-----------------------------------------------------------------
morciniec.ntup_bench.ps.Z

     The n-tuple Classifier: Too good to Ignore  (11 Pages)
         Michal Morciniec and Richard Rohwer

The n-tuple classifier is compared to 23 other methods on 11
large datasets from the European Community StatLog project.

-----------------------------------------------------------------
rohwer.ram_bayes.ps.Z

 Two Bayesian treatments of the n-tuple recognition method (8 pages)
                           Richard Rohwer

 Two probabilistic interpretations of the n-tuple recognition
 method are put forward in order to allow this technique to
 be analysed with the same Bayesian methods used in connection
 with other neural network models.
-----------------------------------------------------------------

Sorry, no hardcopies available.

Three cheers for Jordan Pollack!!!

Richard Rohwer
Dept. of Computer Science and Applied Mathematics
Aston University
Aston Triangle
Birmingham  B4 7ET
ENGLAND

Tel: (44 or 0) (21) 359-3621 x4688
FAX: (44 or 0) (21) 333-6215
rohwerrj at uk.ac.aston.cs


From mike at PARK.BU.EDU  Tue Dec  6 20:51:49 1994
From: mike at PARK.BU.EDU (Michael Cohen)
Date: Tue, 6 Dec 1994 20:51:49 -0500
Subject: WCNN'95 Call for Papers & Meeting Info
Message-ID: <199412070151.UAA25340@cns.bu.edu>

			WORLD CONGRESS ON NEURAL NETWORKS
				RENAISSANCE HOTEL
			      WASHINGTON, DC, USA
				  July 17-21, 1995
CALL FOR PAPERS!
DEADLINE FOR PAPER SUBMISSION:  February 10, 1995.

Registration Fees:
Category		Pre-registration	Pre-registration	On-Site
			prior to		prior to		
			February 10, 1995	June 16, 1995

INNS Member		$170.00			$250.00			$350.00
Non-member**		$270.00			$380.00			$480.00
Student***		$ 85.00			$110.00			$135.00

**Registration fee includes 1995 membership and a one (1) year subscription to 
the Journal Neural Networks.
***Student registration must be accompanied by a letter of verification from 
department chairperson.  Any student registration received with no 
verification letter will be processed at the higher member or non-member fee,  
depending on current membership status.  Copies of student identification cards 
are NOT acceptable.  This also applies to on-site registration.

ORGANIZING COMMITTEE
John G. Taylor, General Chair
Walter J. Freeman
Harold Szu
Rolf Eckmiller
Shun-ichi Amari
David Casasent

INNS OFFICERS				GOVERNING BOARD
President:  Walter J. Freeman		Shun-ichi Amari
President-Elect:  John G. Taylor	James A. Anderson
Past President:  Harold Szu		Andrew Barto
Secretary:  Gail Carpenter		David Casasent
Treasurer:  Judith Dayhoff		Leon Cooper
Executive Director:  R. K. Talley	Rolf Eckmiller
					Kunihiko Fukushima
					Stephen Grossberg
					Mitsuo Kawato
					Christof Koch
					Bart Kosko
					Christoph von der Mallsburg
					Alianna Maren
					Paul Werbos
					Bernard Widrow
					Lotfi A. Zadeh

PROGRAM COMMITTEE
Shun-ichi Amari				James A. Anderson
Etienne Barnard				Andrew R. Barron
Andrew Barto				Theodore Berger
Artie Briggs				Gail Carpenter
David Casasent				Leon Cooper
Judith Dayhoff				Rolf Eckmiller
Jeff Elman				Terrence L. Fine
Francoise Fogelman-Soulie		Walter J. Freeman
Kunihiko Fukushima			Patric Gallinari
Apostolos Georgopoulos			Stephen Grossberg
John B. Hampshire II			Michael Hasselmo
Robert Hecht-Nielsen			Akira Iwata
Jari Kangas				Bert Kappen
Christof Koch				Teuvo Kohonen
Kenneth Kreutz-Delgado			Clifford Lau
Sam Levin				Daniel S. Levine
William B. Levy				Christof von der Malsburg
Alianna Maren				Lina Massone
Lance Optican				Paul Refenes
Jeffrey Sutton				Harold Szu
John G. Taylor				Brian Telfer
Andreas Weigand				Paul Werbos
Hal White				Bernard Widrow
Daniel Wolpert				Kenji Yamanishi
Mona E. Zaghloul

CALL FOR PAPERS:
PAPERS MUST BE RECEIVED BY FEBRUARY 10, 1995.  Authors must submit 
registration payment with papers to be eligible for the early 
registration fee.  A $35 per paper processing fee must be enclosed 
in order for the paper to be refereed.  Please make checks payable 
to INNS and include with submitted paper.  For review purposes, 
please submit six (6) copies (1 original, 5 copies) plus 3 1/2" disk 
(see instructions below), four page limit, in English.  $20 per page 
for papers exceeding (4) pages (do not number pages).  Checks for 
over length charges should be made out to INNS and must be included 
with submitted Paper.  Papers must be on 8 1/2" x 11" white paper 
with 1" margins on all sides, one column format, single spaced, in 
Times or similar type style of 10 points or larger, one side of paper 
only.  FAX's not acceptable.  Centered at the top of first page 
should be complete title, author name(s), affiliation(s), and mailing 
address(es), followed by blank space, abstract (up to 15 lines), and
text.  The following information MUST be included in an accompanying 
cover letter in order for the paper to be reviewed:  Full title of 
paper, corresponding author and presenting author name, address, 
telephone and fax numbers.  Technical Session (see session topics) 
1st and 2nd choices, oral or poster presentation preferred*, 
audio-visual requirements (for oral presentations only).  Papers 
submitted which do not meet these requirements or for which 
insufficient funds are submitted will be returned.

For the first time, the proceedings of the 1995 World Congress on 
Neural Networks will be distributed on CD-ROM.  The CD-ROM Proceedings 
are included in your registration fee.  Please note that the 4 volume 
book of proceedings will not be printed.  Format of 3 1/2" disk for 
CD-ROM:  Once paper is proofed, completed and printed for review, 
reformat the paper in Landscape format, page size 8" x 5" for CD.  
You may include a separate file with 1 paragraph biographical 
information with your name, company, address and telephone number.  
Presenters should submit their papers in one of the following 
Macintosh or Microsoft Windows formats:  Microsoft Word, 
WordPerfect, FrameMaker, Quark or Quark Professional, PageMaker,
Persuasion, ASCII, PowerPoint, Adobe.PDF, Postscript (text, not 
EPS).  Images can be submitted in TIF or PCX format.  If submitting 
in Macintosh format, you should submit a disk containing a Font 
Suitcase of fonts used in your presentationn to ensure proper match 
or enclose a cover letter with paper indicating a list of fonts 
used in your presentation.  Information published on the CD will 
consist of text, in-line and offset mathematical equations,
black and white images, line art and graphics.  By submitting a 
previously unpublished paper, author agrees to the transfer of the
copyright to INNS for the conference proceedings.  All submitted 
papers become the property of INNS.  Papers and disk to be sent to:

WCNN'95, 875 Kings Highway, Suite 200, Woodbury, NJ 08096-3172 USA.

*When to choose a poster presentation:  more than 15 minutes are 
needed; author does not wish to run concurrently with an invited
talk; work is a continuation of previous publication; author seeks 
a lively discussion opportunity or to solicit future collaboration.

PLENARY SPEAKERS:
Daniel L. Alkon, U.S. National Institutes of Health
Shun-ichi Amari, University of Tokyo
Gail Carpenter, Boston University
Walter J. Freeman, University of California, Berkeley
Teuvo Kohonen, Helsinki University of Technology
Harold Szu, Naval Surface Warfare Center
John G. Taylor, King's College London

SESSION TOPICS:
1.	Biological Vision:  Rolf Eckmiller
2.	Machine Vision:  Kunihiko Fukushima, Robert Hecht-Nielsen
3.	Speech and Language:  Jeff Elman
4.	Biological Sensory-Motor Control:  Andrew Barto, Lina Massone
5.	Neurocontrol and Robotics:  Paul Werbos
6.	Supervised Learning:  Andrew R. Barron, Terrence L. Fine
7.	Unsupervised Learning:  Teuvo Kohonen, Francoise Fogelman-Soulie
8.	Pattern Recognition:  David Casasent, Brian Telfer
9.	Prediction and System Identification:  John G. Taylor, Paul Werbos
10.	Cognitive Neuroscience:  James Anderson, Jeffrey Sutton
11.	Links to Cognitive Science & Artificial Intelligence:  Alianna 
	Maren
12.	Signal Processing:  Bernard Widrow
13.	Neurodynamics and Chaos:  Harold Szu,  Mona E. Zaghloul
14.	Hardware Implementation:  Clifford Lau
15.	Associative Memory:  Christoph von der Malsburg
16.	Applications:  Leon Cooper
17.	Circuits and Systems Neuroscience:  Stephen Grossberg, Lance 
	Optican
18.	Mathematical Foundations:  Shun-ichi Amari, D.S. Levine
19.	Evolutionary Computing, Genetic Algorithms:  Judith Dayhoff

SHORT COURSES:
a.	Pattern Recognition and Neural Nets:  David Casasent, Carnegie 
	Mellon University
b.	Modelling Consciousness:  John G. Taylor, King's College London
c.	Neocognitron and the Selective Attention Model:  Kunihiko 
	Fukushima, Osaka University
d.	What are the Differences & the Similarities Among Fuzzy, Neural,
	& Chaotic Systems:  Takeshi Yamakawa, Kyushu Institute of
	Technology
e.	Image Processing & Pattern Recognition by Self-Organizing Neural 
	Networks:  Stephen Grossberg, Boston University
f.	Dynamic Neural Networks:  Signal Processing & Coding:  Judith 
	Dayhoff, University of Maryland
g.	Language and Speech Processing:  Jeff Elman, University of 
	California-San Diego
h.	Introduction to Statistical Theory of Neural Networks:  Shun-ichi 
	Amari, University of Tokyo
i.	Cognitive Network Computation:  James Anderson, Brown University
j.	Biology-Inspired Neural Networks:  From Brain Research to 
	Applications in Technology & Medicine:  Rolf Eckmiller, 
	University of Dusseldorf
k.	Neural Control Systems:  Bernard Widrow, Stanford University
l.	Neural Networks to Advance Intelligent Systems:  Alianna Maren, 
	Accurate Automation Corporation
m.	Reinforcement Learning:  Andrew G. Barto, University of 
	Massachusetts
n.	Advanced Supervised-Learning Algorithms and Applications:  
	Francoise Fogelman-Soulie, SLIGOS
o.	Neural Network & Statistical Methods for Function Estimation:  
	Vladimir Cherkassky, University of Minnesota
p.	Adaptive Resonance Theory:  Gail A. Carpenter, Boston University
q.	What Have We Learned from Experiences of Real World Applications 
	in NN/FS/GA?:  Hideyuki Takagi, Matsushita Elctrical Industrial Co., Ltd.
r.	Fuzzy Function Approximation:  Julie A. Dickerson, University of 
	Southern Califorrnia
s.	Fuzzy Logic and Calculi of Fuzzy Rules and Fuzzy Graphs:  Lofti 
	A. Zadeh, University of California-Berkeley
t.	Overview of Neuroengineering and Supervised Learning:  Paul 
	Werbos, National Science Foundation

INDUSTRIAL ENTERPRISE DAY:
Monday, July 17, 1995
Enterprise Session:
Chair:	Robert Hecht-Nielsen, HNC, Inc.
Industrial Session:
Chair:	Takeshi Yamakawa, Kyushu Institute of Technology

FUZZY NEURAL NETWORKS:
Tuesday, July 18, 1995
Wednesday, July 19, 1995
Co-Chairs:	Bart Kosko, University of Southern California
		Ronald R. Yager, Iona College

SPECIAL SESSIONS:
Neural Network Applications in the Electrical Utility Industry
Biomedical Applications & Imaging/Computer Aided Diagnosis in 
Medical Imaging
Statistics and Neural Networks
Dynamical Systems in Financial Engineering
Mind, Brain and Consciousness
Physics and Neural Networks
Biological Neural Networks

To obtain additional information (complete registration brochure, 
registration and hotel forms) contact WCNN'95, 875 Kings Highway,
Suite 200, Woodbury, New Jersey 08096-3172 USA, Tele:  (609)845-
1720; Fax:  (609)853-0411; e-mail:  74577.504 at compuserve.com




From jbower at smaug.bbb.caltech.edu  Wed Dec  7 17:13:15 1994
From: jbower at smaug.bbb.caltech.edu (jbower@smaug.bbb.caltech.edu)
Date: Wed, 7 Dec 94 14:13:15 PST
Subject: Call for papers -- CNS*95
Message-ID: <mailman.714.1149591321.29955.connectionists@cs.cmu.edu>

                          CALL FOR PAPERS                        

          Fourth Annual Computational Neuroscience Meeting 
                             CNS*95 
                       July 11 - 15, 1995
                       Monterey, California
                         ................
               DEADLINE FOR SUMMARIES AND ABSTRACTS: 

                   **>> January 25, 1995 <<**
                        ^^^^^^^^^^^^^^^^
        This is the fourth annual meeting of an interdisciplinary
conference intended to address the broad range of research approaches and
issues involved in the field of computational neuroscience.  The last three
annual meetings, in San Francisco (CNS*92), Washington, DC (CNS*93), and
Monterey, California (CNS*94) brought experimental and theoretical
neurobiologists along with engineers, computer scientists, cognitive
scientists, physicists, and mathematicians together to consider the
functioning of biological nervous  systems.  Peer reviewed papers were
presented on a range of subjects related to understanding how nervous
systems compute.

        As in previous years, the meeting will equally emphasize
experimental, model-based, and more abstract theoretical  approaches to
understanding neurobiological computation.  The meeting in 1995 will again
take place at the Monterey Doubletree Hotel and include plenary,
contributed, and poster  sessions.  There will be no parallel sessions and
the full text of presented papers will be published in a proceedings
volume.  The last day of the meeting will be devoted to a series of
informal workshops focused on current issues in computational neuroscience.
 Student Travel funds and Child Day Care will be available.

                       SUBMISSION INSTRUCTIONS:

      With this announcement we solicit the submission of presented papers.
 All papers will be refereed.  Authors should send original research
contributions in the form of a 1000-word (or less) summary and a separate
single page 50-100 word abstract clearly stating their results.  Summaries
are for program committee use only.  Abstracts will be published in the
conference program.  At the bottom of each abstract page and on the first
summary page,  indicate preference for oral or poster presentation and
specify at least one appropriate category and theme from the following 
list:
 
 Presentation categories: 
    A. Theory and Analysis 
    B. Modeling and Simulation
    C. Experimental 
    D. Tools and Techniques


Themes: 
   A. Development 
   B. Cell Biology 
   C. Excitable Membranes and Synaptic   
        Mechanisms 
   D. Neurotransmitters, Modulators, 
        Receptors 
   E. Sensory Systems
       1. Somatosensory 
       2. Visual
       3. Auditory 
       4. Olfactory 
       5. Other systems
   F. Motor Systems and Sensory Motor
     Integration 
   G. Learning and Memory
   H. Behavior 
   I. Cognitive  
   J. Disease     

 
        Include addresses of all authors on the front of the summary and
the abstract including the E-mail address for EACH author. Indicate on the
front of the summary to which author correspondence should be addressed. 
Program committee decisions will be sent to the correspondence author only.
Submissions will not be considered if they lack category information,
separate abstract sheets,  author addresses, or are late.

Submissions can be made by surface mail ONLY by sending 6 copies of the
abstract and summary to:        

        CNS*95 Submissions
        Division of Biology    216-76
        Caltech    
        Pasadena, CA  91125



ADDITIONAL INFORMATION can be obtained by: 

        o Using our on-line WWW information and registration server, URL of:
            http://www.bbb.caltech.edu/cns95.html

        o ftp-ing to our ftp site.
            yourhost% ftp 131.215.137.69
            Name (131.215.137.69:<yourname>): ftp
            Password:  yourname at yourhost.yourside.yourdomain
            ftp> cd cns95
            ftp> ls

        o Sending Email to:
            cns95-registration-info at smaug.bbb.caltech.edu   


CNS*94 ORGANIZING COMMITTEE:

    Co-meeting chair / logistics -   John Miller, UC Berkeley
    Co-meeting chair / program  -    Jim Bower, Caltech

  Program committee:
      Gwen Jacobs,  University of California,  Berkeley
      Catherine Carr, University of Maryland, College Park
      Dennis Glanzman, NIMH/NIH
      Nancy Kopel, Boston University
      Christiane Linster, ESPCI, Paris, France
      Philip Ulinski,  University of Chicago
      Charles Wilson,  University of Tennessee, Memphis

  Regional Organizers:
                          
       Europe- Erik DeSchutter (Belgium)
       Middle East - Idan Segev (Jerusalem)
       Down Under - Mike Paulin (New Zealand)
       South America - Renato Sabbatini (Brazil)
       Asia - Zhaoping Li (Hong Kong)



           ***************************************
                       James M. Bower
                     Division of Biology
                     Mail code:  216-76
                           Caltech
                     Pasadena, CA 91125
                      (818) 395-6817
                      (818) 449-0679 FAX

      NCSA Mosaic laboratory address:  http://www.bbb.caltech.edu/bowerlab
      NCSA Mosaic address for GENESIS:  http://www.bbb.caltech.edu/GENESIS




From lars at ida.his.se  Wed Dec  7 02:31:17 1994
From: lars at ida.his.se (Lars Niklasson)
Date: Wed, 7 Dec 94 08:31:17 +0100
Subject: Lectureship 
Message-ID: <9412070731.AA00135@mhost.ida.his.se>

The following lectureship is available for a connectionist 
with a background in Lingusitics.

===========================================================

Lectureship in linguistics, ref. no. 276-94-40 
University of Skoevde, Sweden.

Deadline for application: Jan. 20th 1995.

The appointment includes both research and lecturing duties.
Applicants  should  posses  a general competence in linguis-
tics, but also be prepared to work within  the  border-areas
of  computer  science and cognitive science, which are areas
where the University is planning to develop  both  education
and  a research competence with special emphasis on linguis-
tics. It is therefore considered a merit  if  the  applicant
has  skills  within  traditional  artificial intelligence or
connectionism. The lectureship will belong to the Department
of  modern  languages,  but will involve lecturing duties at
the Department of computer science.

The University of Skoevde is one of Sweden's  youngest,  but
it  is undergoing a dynamical development. The main focus of
all the University activities is towards the private sector,
industry  and  international  trade,  especially the use and
development of tools for information technology.  Six  areas
of  competence  have been established; engineering, computer
science, cognitive science, economics, language as  well  as
media and arts.

Currently, the education programmes  of  the  Department  of
Modern  Languages include; English, French, Spanish and Ger-
man. In addition, the Department of computer science  offers
a number of 3-year programmes in computer science and cogni-
tive science, as well as a  research  oriented  master  pro-
gramme.  Research is conducted in connectionism, distributed
real-time databases and active databases.

For more information please  contact:  
The vice-chancellor's office  
Lars-Erik Johansson

Email: Lars-Erik.Johansson at sta.his.se






From pollack at cs.brandeis.edu  Wed Dec  7 17:12:09 1994
From: pollack at cs.brandeis.edu (Jordan Pollack)
Date: Wed, 7 Dec 1994 17:12:09 -0500
Subject: compneuro asst prof job
Message-ID: <199412072212.RAA02297@onyx.cs.brandeis.edu>

I recently moved to Brandeis University, near Boston, where there is a
lovely new building housing a new center for complex systems, which
focuses on the brain, the mind, and computation. The center houses
biologists, biochemists, physicists, psychologists, linguists, and
computer scientists.

There is a faculty search which is relevant to the list:

-------

The Volen National Center for Complex Systems at Brandeis University seeks
candidates for a tenure-track assistant professorship to participate in
establishing a Center for Theoretical Neuroscience funded by a grant from
the Alfred P. Sloan Foundation. We are especially interested in theorists
with strong interests and demonstrated commitment to research relevant to
neuroscience.  Candidates are expected to have Ph.D.s (or the equivalent) in
Mathematics, Physics, Computer Science or another quantitative discipline.
The successful candidate will be appointed in the appropriate academic
department/s.  The position carries a reduced teaching load and summer
salary for three years.  Prospective candidates should send a curriculum
vitae, statement of research interests, and arrange for three letters of
recommendation to be sent directly to: Search Committee,
Theoretical/Computational Neuroscientist, Volen Center, Brandeis University,
Waltham, MA 02254. Applications from minority and women are especially
welcome. Consideration of completed applications will start after January 1,
1995, and will continue until the position is filled, although candidates
are strongly encouraged to submit completed applications as soon as
possible.

----

The likeliest candidates for the faculty position will have a body of
work in neuroscience modeling, rather than in neural models of
cognition or engineering. However, there are also several postdocs
available on the same Sloan grant, for theorists or modelers
who want to cross into experimental neuroscience. Contact Eve
Marder at binah.cc.brandeis.edu, Larry Abbott at psy.ox.ac.uk or John
Lisman at binah.cc.brandeis.edu for further info.

Jordan Pollack                           Associate Professor
Computer Science Department              Center for Complex Systems
Brandeis University                      Phone: (617) 736-2713/fax 2741
Waltham, MA 02254                        email: pollack at cs.brandeis.edu

PS Neuroprose is staying at OSU for a while, just with delays in moving files.









From pollack at cs.brandeis.edu  Thu Dec  8 11:51:23 1994
From: pollack at cs.brandeis.edu (Jordan Pollack)
Date: Thu, 8 Dec 1994 11:51:23 -0500
Subject: FLAME:  WCNN'95 - science in the mud pits
Message-ID: <199412081651.LAA01963@garnet.cs.brandeis.edu>

I kept quiet about the first IEEE NN conference in 1987, where every
paper was accepted, and sorted by how well the author's name was
recognized by the local committee, a feast of surplus dollars were
collected in hyperinflated registration fees.

I just never submitted another paper. But these policies are sickening:

>Authors must submit 
>registration payment with papers to be eligible for the early 
>registration fee...Registration fee includes 1995 membership and a 
>one (1) year subscription to  the Journal Neural Networks.
>A $35 per paper processing fee must be enclosed 
>in order for the paper to be refereed.

Reading fees are charged by agents for aspiring science fiction
writers. Linking registration to submission is nothing more than an
admission that all papers are to be accepted.  And a fee of $480 is
paying a lot more than membership, subscription, and a few days of
coffee breaks!

Also, since the size of the intersection of the two sets
   |ORGANIZING COMMITTEE & PLENARY SPEAKERS| 
is 4 instead of 0, I guess a pentium chip was used.

WHERE are the ethical governors of this scientific society?

Jordan Pollack                           Associate Professor
Computer Science Department              Center for Complex Systems
Brandeis University                      Phone: (617) 736-2713/fax 2741
Waltham, MA 02254                        email: pollack at cs.brandeis.edu


From carlos at cenoli1.ulb.ac.be  Thu Dec  8 16:15:38 1994
From: carlos at cenoli1.ulb.ac.be (carlos@cenoli1.ulb.ac.be)
Date: Thu, 8 Dec 94 16:15:38 MET
Subject: paper available: dynamical computation with chaos
Message-ID: <9412081515.AA01490@cenoli5.ulb.ac.be>


FTP-host: archive.cis.ohio-state.edu
FTP-filename: /pub/neuroprose/babloyantz.comput_chaos.ps.Z

The file babloyantz.comput_chaos.ps.Z is now available for copying
from the Neuroprose repository. This is a 13-page long paper with
5 figures. Size is 204239 bytes compressed and 585704 uncompressed.
Sorry, hardcopies not available.

    Computation with chaos: A paradigm for cortical activity

                  A. Babloyantz and C. Louren\c{c}o

   Service de Chimie-Physique, Universit\'{e} Libre de Bruxelles
        CP 231 - Campus Plaine, Boulevard du Triomphe
               B-1050~Bruxelles, Belgium
            e-mail: carlos at cenoli.ulb.ac.be

Abstract: A device comprising two interconnected networks of
oscillators exhibiting spatiotemporal chaos is considered.
An external cue stabilizes input specific unstable periodic orbits
of the first network thus creating an ``attentive'' state.
Only in this state is the device able to perform pattern discrimination
and motion detection. We discuss the relevance of the procedure to the
information processing of the brain.

This paper appeared in Proc. Natl. Acad. Sci. USA, Vol. 91, p. 9027,
13 September 1994.

Thanks to Jordan Pollack for maintaining the archive.

          Carlos Louren\c{c}o

-------------------------------------------------------------------------

    *** Indonesians persist in the racial cleansing of East Timor. ***
       *** Torture and killing of innocents happen daily. ***


From mdavies at psy.ox.ac.uk  Wed Dec  7 11:24:21 1994
From: mdavies at psy.ox.ac.uk (Martin Davies)
Date: Wed, 7 Dec 1994 16:24:21 +0000
Subject: Euro-SPP '95
Message-ID: <9412071624.AA21454@Mac8>

*************************************************************************

EUROPEAN SOCIETY FOR PHILOSOPHY AND SOCIETY

Fourth Annual Meeting
St. Catherine's College, Oxford
Wednesday 30 August - Friday 1 September, 1995

*************************************************************************

FIRST ANNOUNCEMENT AND CALL FOR PAPERS

The Fourth Annual Meeting of the Euro-SPP will begin at 11.30 am on 
Wednesday 30 August and will end at 5.30 pm on Friday 1 September.  

Themes for Invited Symposia include: emotion, attention, artifical life, 
and brain imaging.  The conference will be held in St. Catherine's College, 
Oxford, where accommodation will be available.

We expect to be able to offer an accommodation and meals package for the 
period from Wednesday morning until Friday afternoon for 108 pounds.  In
addition, bed and breakfast accommodation will be available for the Tuesday 
night before the conference, and for the Friday and Saturday nights after the
conference, at a cost of 28 pounds per night.  A limited number of superior 
rooms with private bath will be available at a higher rate.

*************************************************************************

For further information about local arrangements, email:  espp95 at psy.ox.ac.uk.

*************************************************************************

The Society welcomes submitted papers and posters for this meeting.  

Submitted papers and posters are refereed and selected on the basis of quality 
and relevance to both psychologists and philosophers.

Submitted Papers:  Papers should not exceed a length of 30 minutes (about 
12 double-spaced pages).  The full text should be submitted, along with a 300 
word abstract.

Poster Presentations:  Proposals for poster presentations should consist of a 
500 word abstract.  Unless authors indicate otherwise, submitted papers that 
we are not able to accept will also be considered for poster presentation.

The deadline for submission of both submitted papers and poster presentations 
is 20 January 1995.  Please send three copies to:

	Professor Beatrice de Gelder
	Department of Psychology
	Tilburg University
	5000 LE Tilburg
	The Netherlands
or:
	Professor Christopher Peacocke
	Magdalen College
	Oxford  OX1 4AU
	UK

*************************************************************************

For information about membership of the Euro-SPP, email:  espp at kub.nl.

*************************************************************************








From marks at u.washington.edu  Fri Dec  9 20:37:22 1994
From: marks at u.washington.edu (Robert Marks)
Date: Fri, 9 Dec 94 17:37:22 -0800
Subject: Russian Symposium
Message-ID: <9412100137.AA25592@carson.u.washington.edu>


              The 2-nd International Symposium
           on Neuroinformatics and Neurocomputers
                   Rostov-on-Don, RUSSIA
                   September 20-23, 1995

     Organized by Russian Neural Network Society (RNNS)

                          and

 A.B. Kogan Research Institute for Neurocybernetics (KRINC)

                   in co-operation with

     Institute of Electrical and Electronics Enginieers
             Neural Networks Council (IEEE NNC)


                   First Call for Papers

Research in Neuroinformatics and Neurocomputing continued in
Russia after  the research  was deflated  in the west in the
1970's.   The research sophistication in neural networks, as
a   result,  is   quite   advanced   in   Russia.  The first
international  RNNS/IEEE  Symposium, held  in October  1992,
proved  to  be  a  highly  successful  forum  for  a diverse
international  interchange  of   fresh  and  novel  research
results.    The    second   International    Symposium    on
Neuroinformatics   and   Neurocomputers   is   built on this
remarkable success.

The symposium  focus is  on the  neuroscience,  mathematics,
physics,  engineering  and  design  of  neuroinformatic  and
neurocomputing systems.

Rostov-on-Don, the  location of the Symposium, is about 1000
km south  of Moscow  on the  scenic Don  river.   The Don is
commonly identified  as the  boundary between the continents
of Europe  and Asia.   Rostov  is the home of the A.B. Kogan
Research Institute  for  Neurocybernetics  at  Rostov  State
University -  one of  the premier  neural  network  research
centers in Russia.

Papers for  the Symposium  should be  sent  in  CAMERA-READY
FORM, NOT  EXEEDING 8  PAGES in A4  format  to   the Program
Committee Co-Chair Alexander A. Frolov.   Two copies  of the
paper should  be submitted.

The deadline for submission is 15 MARCH, 1995.

Notification of acceptance will be sent on or before 15 May,
1995.




                    SYMPOSIUM COMMITTEE

GENERAL CHAIR

     Witali L. Dunin-Barkowski, Dr. Sci.,
     The 2-nd International Symposium
     on Neuroinformatics and Neurocomputers,
     Symposium Chair
     President of the Russian Neural Network Society,
     A.B. Kogan Research Institute for Neurocybernetics
     Rostov State University
     194/1 Stachka avenue,
     344104, Rostov-on-Don, Russia
     Tel: +7-8632-28-0588,
     Fax: +7-8632-28-0367
     E-mail: wldb at krinc.rostov-na-donu.su


PROGRAM COMMITTEE CO-CHAIRS

     Professor Alexander A. Frolov,
     The 2-nd International Symposium on Neuroinformatics and
        Neurocomputers,
     Program Co-Chair
     5a Butlerov str.
     Higher Nervous Activity and Neurophysiology Institute
     Russian Academy of Science
     117220, Moscow, RUSSIA.

     Professor Robert J. Marks II
     Program Co-Chair
     The 2-nd International Symposium on Neuroinformatics and
        Neurocomputers
     University of Washington
     Department of Electrical Engineering
     c/o 1131 199th Street S.W., Suite N
     Lynnwood, WA 98036-7138
     WA, USA.

Other information is available from the Symposium Committee.





From john at dcs.rhbnc.ac.uk  Sat Dec 10 07:18:37 1994
From: john at dcs.rhbnc.ac.uk (john@dcs.rhbnc.ac.uk)
Date: Sat, 10 Dec 94 12:18:37 +0000
Subject: Technical Report Series in Neural and Computational Learning
Message-ID: <28465.9412101218@platon.cs.rhbnc.ac.uk>


The European Community ESPRIT Working Group in Neural and Computational 
       Learning Theory (NeuroCOLT): one new report available

----------------------------------------
NeuroCOLT Technical Report NC-TR-94-011:
----------------------------------------
Valid Generalisation from Approximate Interpolation

by  Martin Anthony, Department of Mathematics, The London School of Economics
    Peter Bartlett, Research School of Information Sciences and  Engineering,
       Australian National University
    Yuval Ishai, Department of Computer Science, Technion
    John Shawe-Taylor, Department of Computer Science, Royal Holloway, 
       University of London

Abstract:
Let $\H$ and $\C$ be  sets of functions from domain $X$ to the reals. We say
that $\H$ validly generalises $\C$ from approximate interpolation if and only if
for each $\eta>0$ and $\epsilon, \delta \in (0,1)$ there is a number
$m_0(\eta,\epsilon, \delta)$ such that for any function $t \in \C$ and any
probability distribution $P$ on $X$,  if $m \ge m_0$ then with $P^m$-probability
at least $1-\delta$, a sample $\vx =(x_1, x_2, \dots, x_m) \in X^m$ satisfies
$$\forall h \in \H, \,
|h(x_i) - t(x_i)|< \eta, \,(1 \le i \le m) \Longrightarrow
 P(\{x: |h(x) -t(x)| \ge \eta\}) < \epsilon.$$
We find conditions that are necessary and sufficient for $\H$ to validly
generalise $\C$ from approximate interpolation, and we obtain bounds on the
sample length $m_0(\eta, \epsilon, \delta)$ in terms of various parameters
describing the expressive power of $\H$.

----------------------------------------
NeuroCOLT Technical Report NC-TR-94-022:
----------------------------------------
Sample Sizes for Sigmoidal Neural Networks
by John Shawe-Taylor, Department of Computer Science, Royal Holloway
       University of London

Abstract:
This paper applies the theory of Probably Approximately Correct (PAC)
learning to feedforward neural networks with sigmoidal activation
functions.  Despite the best known upper bound on the VC dimension of
such networks being $O((WN)^2)$, for $W$ parameters and $N$
computational nodes, it is shown that the asymptotic bound on the sample
size required for learning with increasing accuracy $1 - \epsilon$ and
decreasing probability of failure $\delta$ is
$$O((1/\epsilon)(W\log(1/\epsilon) + (WN)^2 + \log(1/\delta)).$$ 
For practical values of $\epsilon$ and $\delta$ the formula obtained for
the sample sizes is a factor $2\log(2e/\epsilon)$ smaller than a naive
use of the VC dimension result would give. Similar results are obtained
for learning where the hypothesis is only guaranteed to correctly
classify a given proportion of the training sample. The results are
formulated in general terms and show that for many learning classes
defined by smooth functions thresholded at the output, the sample size
for a class with VC-dimension $d$ and $\ell$ parameters is
$O((1/\epsilon)(\ell\log(1/\epsilon) + o(\log(1/\epsilon))d +
\log(1/\delta))$.  

-----------------------
The Report NC-TR-94-011 can be accessed and printed as follows 

% ftp cscx.cs.rhbnc.ac.uk  (134.219.200.45)
Name: anonymous
password: your full email address
ftp> cd pub/neurocolt/tech_reports
ftp> binary
ftp> get nc-tr-94-011.ps.Z
ftp> bye
% zcat nc-tr-94-011.ps.Z | lpr -l

Similarly for the other technical reports.

Uncompressed versions of the postscript files have also been
left for anyone not having an uncompress facility.

A full list of the currently available Technical Reports in the Series is held in a file
`abstracts' in the same directory.

Best wishes
John Shawe-Taylor




From cnna at tce.ing.uniroma1.it  Mon Dec 12 09:54:22 1994
From: cnna at tce.ing.uniroma1.it (cnna@tce.ing.uniroma1.it)
Date: Mon, 12 Dec 1994 15:54:22 +0100
Subject: program CNNA-94
Message-ID: <9412121454.AA14007@tce.ing.uniroma1.it>


(Multiple posting - please excuse us if you receive more than a copy)

THIRD IEEE INTERNATIONAL WORKSHOP ON CELLULAR NEURAL NETWORKS AND THEIR 
APPLICATIONS (CNNA-94)
ROME, ITALY, DEC. 18-21, 1994

PRELIMINARY PROGRAM

SUNDAY, DECEMBER 18, 1994 

17.30 Inaugural Session (Hall 1)

17.30 Welcome address
G. Orlandi, Dean, Faculty of Engineering
V. Cimagalli, Chairman, CNNA-94

17.45 Opening address
C. Lau

18.05 Inaugural lecture:
L.O. Chua
The CNN Universal Chip: Dawn of a New Computer Paradigm

19.30 Welcome Cocktail (Room of the Frescoes)

MONDAY, DECEMBER 19, 1994

9.00 Session 1: Theory I (Hall of the Cloister - chairman: L.O. Chua)

9.00 Invited review paper:
T. Roska
Analogic Algorithms Running on the CNN Universal Machine

9.30 G. Yang, T. Yang, L.-B. Yang
On Unconditional Stability of the General Delayed Cellular Neural 
Networks 
9.45 S. Arik, V. Tavsanoglu
A Weaker Condition for the Stability of Nonsymmetric CNNs
10.00 M. P. Joy, V. Tavsanoglu
Circulant Matrices and the Stability Theory of CNNs
10.15 B.E. Shi, S. Wendsche, T. Roska, L.O. Chua
Random Variations in CNN Templates: Theoretical Models and Empirical 
Studies

10.30 Coffee Break (Room of the Frescoes)

11.00 Session 2: Connections with Neurophysiology (Hall of the Cloister - 
chairman: T. Roska)

11.00 Invited lecture:
F. Werblin, A. Jacobs 
Using CNN to Unravel Space-Time Processing in the Vertebrate Retina

11.45 Invited lecture:
J. Hmori
Synaptic Organization of the Thalamic Visual Center (LGN) of Mammals as a 
Basis of CNN Model of Subcortical Visual Processing

12.30 K. Lotz, Z. Vidnynszky, T. Roska, J. Vandewalle, J. Hmori, A. 
Jacobs, F. Werblin
Some Cortical Spiking Neuron Models Using CNN
12.45 T.W. Berger, B.J. Sheu, R. H.-J. Tsai
Analog  VLSI Implementation of a Nonlinear Systems Model of  the 
Hippocampal Brain Region
13.00 A. Jacobs, T. Roska, F. Werblin
Techniques for Constructing Physiologically Motivated Neuromorphic Models 
in CNN

13.15 Lunch (Room of the Frescoes)

14.45 Session 3: Hardware Implementations I (Hall of the Cloister - 
chairman: A. Rodrguez-Vzquez)

14.45 Invited review paper:
A. Rodrguez-Vzquez, R. Domnguez-Castro, S. Espejo
Design of CNN Universal Chips: Trends and Obstacles

15.15 J.M. Cruz, L.O. Chua, T. Roska
A Fast, Complex and Efficient Test Implementation of the CNN Universal 
Machine
15.30 F. Sargeni, V. Bonaiuto
High Performance Digitally Programmable CNN Chip with Discrete Templates
15.45 A. Paasio, A. Dawidziuk, K. Halonen, V. Porra
Digitally Controllable Weights in Current Mode Cellular Neural Networks
16.00 D. Lm, G.S. Moschytz
A Programmable, Modular CNN Cell
16.15 M.-D. Doan, R. Chakrabaty, M. Heidenreich, M. Glesner, S. Cheung
Realisation of a Digital Cellular Neural Network for Image Processing
16.30 R. Domnguez-Castro, S. Espejo, A. Rodrguez-Vzquez, R. Carmona
A CNN Universal Chip in CMOS Technology

16.45 Coffee Break (Room of the Frescoes)

17.15 Session 4: Theory II (Hall of the Cloister - chairman: R.-W. Liu)

17.15 R.-W. Liu, Y.-F. Huang, X.-T. Ling
A Novel Approach to the Convergence of Neural Networks for Signal 
Processing
17.30 E. Pessa, M.P. Penna
Local and Global Connectivity in Neuronic Cellular Automata
17.45 X.-Z. Huang, T. Yang, L.-B. Yang
On Stability of the Time-Variant Delayed Cellular Neural Networks
18.00 J.J. Szczyrek, S. Jankowski
A Class of Asymmetrical Templates in Cellular Neural Networks
18.15 P.P. Civalleri, M. Gilli
A Topological description of the State Space of a Cellular Neural Network
18.30 M. Tanaka, T. Watanabe
Cooperative and Competitive Cellular Neural Networks

TUESDAY, DECEMBER 20, 1994

9.15 Session 5: Learning I (Hall of the Cloister - chairman: J.A. Nossek)

9.15 Invited review paper:
J.A. Nossek
Design and Learning with Cellular Neural Networks

9.45 I. Fajfar, F. Bratkovic
Statistical Design Using Variable Parameter Variances and Application to 
Cellular Neural Networks
10.00 N.N. Aizenberg, I.N. Aizenberg
CNN-like Networks Based on Multi-Valued and Universal Binary Neurons: 
Learning and Application to Image Processing
10.15 W. Utschick, J.A. Nossek
Computational Learning Theory Applied to Discrete-Time Cellular Neural 
Networks
10.30 H. Magnussen, J.A. Nossek
Global Learning Algorithms for Discrete-Time Cellular Neural Networks

10.45 Coffee Break (Room of the Frescoes)

11.15 Session 6: Learning II (Hall of the Cloister - chairman: J. 
Vandewalle)

11.15 H. Magnussen, G. Papoutsis, J.A. Nossek
Continuation-Based Learning Algorithm for Discrete-Time Cellular Neural 
Networks
11.30 C. Gzelis, S. Karamahmut
Recurrent Perceptron Learning Algorithm for Completely Stable Cellular 
Neural Networks
11.45 A.J. Schuler, M. Brabec, D. Schubel, J.A. Nossek
Hardware-Oriented Learning for Cellular Neural Networks
12.00 F. Dellaert, J. Vandewalle
Automatic Design of Cellular Neural Networks by Means of Genetic 
Algorithms: Finding a Feature Detector
12.15 H. Mizutani
A New Learning Method for Multilayered Cellular Neural Networks

12.30 Lunch (Room of the Frescoes)

14.00 Panel Discussion: Trends and Applications of the CNN Paradigm (Hall 
of the Cloister - chairman: L.O. Chua)
Panelists: L.O. Chua, V. Cimagalli, J.A. Nossek, T. Roska, A. Rodrguez-
Vzquez, J. Vandewalle

15.30 Coffee Break (Room of the Frescoes)

16.00 Session 7: Applications I (Hall of the Cloister - chairman: N.N. 
Aizenberg)

16.00 N.N. Aizenberg, I.N. Aizenberg, T.P. Belikova
Extraction and Localization of Important Features on Grey-Scale Images: 
Implementation on the CNN
16.15 K. Slot
Large-Neighborhood Templates Implementation in Discrete-Time CNN 
Universal 
Machine with a Nearest-Neighbor Connection Pattern
16.30 J. Pineda de Gyvez
XCNN: A Software Package for Color Image Processing
16.45 B.E. Shi
Order Statistic Filtering with Cellular Neural Networks
17.00 L.-B. Yang, T. Yang, B.-S. Chen
Moving Point Target Detection Using Cellular Neural Networks
17.15 X.-P. Yang, T. Yang, L.-B. Yang
Extracting Focused Object from Defocused Background Using Cellular Neural 
Networks
17.30 M. Balsi, N. Racina
Automatic Recognition of Train Tail Signs Using CNNs
17.45 A. Kellner, H. Magnussen, J.A. Nossek
Texture Classification, Texture Segmentation and Text Segmentation with 
Discrete-Time Cellular Neural Networks

16.00 Session 8: Applications II (Hall 17 - chairman: S. Jankowski)

16.00 P.L. Venetianer, P. Szolgay, K.R. Crounse, T. Roska, L.O. Chua
Analog Combinatorics and Cellular Automata-Key Algorithms and Layout 
Design
16.15 . Zarndy, T. Roska, Gy. Liszka, J. Hegyesi, L. Kk, Cs. Rekeczky
Design of Analogic CNN Algorithms for Mammogram Analysis
16.30 P. Szolgay, Gy. Erss, A. Katona, . Kiss
An Experimental System for Path Tracking of a Robot Using a 16*16 
Connected Component Detector CNN Chip with Direct Optical Input
16.45 T. Kozek, T. Roska
A Double Time-Scale CNN for Solving 2-D Navier-Stokes Equations
17.00 . Zarndy, F. Werblin, T. Roska, L.O. Chua
Novel Types of Analogic CNN Algorithms for Recognizing Bank-Notes
17.15 B.J. Sheu, Sa H. Bang, W.-C. Fang
Optimal Solutions of Selected Cellular Neural Network Applications by the 
Hardware Annealing Method
17.30 B. Siemiatkowska
Cellular Neural Network for Mobile Robot Navigation
17.45 A. Murgu
Distributed Neural Control for Markov Decision Processes in Hierarchic 
Communication Networks

20.00 Banquet (Restaurant "4 Colonne" - via della Posta Vecchia, 4)

WEDNESDAY, DECEMBER 21, 1994

9.00 Session 9: Spatio-Temporal Dynamics I (Hall of the Cloister - 
chairman: V.D. Shalfeev)

9.00 Invited review paper
V.D. Shalfeev, G.V. Osipov
Spatio-Temporal Phenomena in Cellular Neural Networks

9.30 C.-M. Yang, T. Yang, K.-Y. Zhang
Chaos in Discrete Time Cellular Neural Networks
9.45 R. Dogaru, A.T. Murgan, D. Ioan
Robust Oscillations and Bifurcations in Cellular Neural Networks
10.00 H. Chen, M.-D. Dai, X.-Y. Wu
Bifurcation and Chaos in Discrete-Time Cellular Neural Networks
10.15 M.J. Ogorzalek, A. Dabrowski, W. Dabrowski
Hyperchaos, Clustering and Cooperative Phenomena in CNN Arrays Composed 
of Chaotic Circuits

10.30 Coffee Break (Room of the Frescoes)

11.00 Session 10: Spatio-Temporal Dynamics II (Hall of the Cloister - 
chairman: M. Hasler)

11.00 Invited review paper:
P. Thiran, M. Hasler
Information Processing Using Stable and Unstable Oscillations: A Tutorial

11.30 P. Szolgay, G. Vrs
Transient Response Computation of a Mechanical Vibrating System Using 
Cellular Neural Networks
11.45 P.P. Civalleri, M. Gilli
Propagation Phenomena in Cellular Neural Networks
12.00 S. Jankowski, A. Londei,  C. Mazur, A. Lozowski 
Synchronization Phenomena in 2D Chaotic CNN

12.15 Poster Session (Hall of the Cloister)

S. Jankowski, R. Wanczuk   
CNN models of complex pattern formation in excitable media
Z. Galias, J.A. Nossek
Control of a Real Chaotic Cellular Neural Network
A. Piovaccari, G. Setti
A Versatile CMOS Building Block for Fully Analogically-ProgrammableVLSI 
Cellular Neural Networks
P. Thiran, G. Setti
An Approach to Local Diffusion and Global Propagation in 1-dim. Cellular 
Neural Networks
J. Kowalski, K. Slot, T. Kacprzak
A CMOS Current-Mode VLSI Implementation of Cellular Neural Network for an 
Image Objects Area Estimation
W.J. Jansen, R. van Drunen, L. Spaanenburg, J.A.G. Nijhuis
The AD2 Microcontroller Extension for Artificial Neural Networks
C.-K. Pham, M. Tanaka
A Novel Chaos Generator Employing CMOS Inverter for Cellular Neural 
Networks

13.15 Lunch (Room of the Frescoes)

14.45 Session 11: Hardware Implementations II (Hall of the Cloister - 
chairman: J.L. Huertas)

14.45 R. Beccherelli, G. de Cesare, F. Palma
Towards an Hydrogenated Amorphous Silicon Phototransistor Cellular Neural 
Network
15.00 A. Sani, S. Graffi, G. Masetti, G. Setti
Design of CMOS Cellular Neural Networks Operating at Several Supply 
Voltages
15.15 M. Russell Grimaila, J. Pineda de Gyvez
A Macromodel Fault Generator for Cellular Neural Networks
15.30 P. Kinget, M. Steyaert
Evaluation of CNN Template Robustness Towards VLSI Implementation
15.45 B.J. Sheu, Sa H. Bang, W.-C. Fang
Analog VLSI Design of Cellular Neural Networks with Annealing Ability
16.00 L. Raffo, S.P. Sabatini, G.M. Bisio
A Reconfigurable Architecture Mapping Multilayer CNN Paradigms

14.45: Session 12: Applications III (Hall 17 - chairman: J. Herault)

14.45 G. Adorni, V. DAndrea, G. Destri
A Massively Parallel Approach to Cellular Neural Networks Image 
Processing
15.00 M. Coli, P. Palazzari, R. Rughi
Use of the CNN Dynamic to Associate Two Points with Different 
Quantization Grains in the State Space
15.15 M. Csapodi, L. Nemes, G. Tth, T. Roska, A. Radvnyi
Some Novel Analogic CNN Algorithms for Object Rotation, 3D Interpolation-
Approximation, and a "Door-in-a-Floor" Problem
15.30 H. Harrer, P.L. Venetianer, J.A. Nossek, T. Roska, L.O. Chua 
Some Examples of Preprocessing Analog Images with Discrete-Time Cellular 
Neural Networks
15.45 A.G. Radvnyi
Solution of Stereo Correspondence in Real Scene: an Analogic CNN 
Algorithm
16.00 J.P. Miller,  K.R. Crounse, T. Sziranyi,  L. Nemes,  L.O. Chua, T. 
Roska
Deblurring of Images by Cellular Neural Networks with Applications to 
Microscopy

16.15 Coffee Break (Room of the Frescoes)

16.45 Session 13: Hardware Implementations III (Hall of the Cloister - 
chairman: M. Salerno)

16.45 T. Roska, P. Szolgay, . Zarndy, P.L. Venetianer, A. Radvnyi, T. 
Szirnyi
On a CNN Chip-Prototyping System
17.00 M. Balsi, V. Cimagalli, I. Ciancaglioni, F. Galluzzi
Optoelectronic Cellular Neural Network Based on Amorphous Silicon Thin 
Film Technology
17.15 S. Espejo, R. Domnguez-Castro, A. Rodrguez-Vzquez, R. Carmona
Weight-Control Strategy for Programmable CNN Chips
17.30 S. Espejo, A. Rodrguez-Vzquez, R. Domnguez-Castro, R. Carmona
Convergence and Stability of the FSR CNN Model
17.45 R. Domnguez-Castro, S. Espejo, A. Rodrguez-Vzquez, I. Garca-
Vargas, J.F. Ramos, R. Carmona
SIRENA: A Simulation Environment for CNNs

16.45 Session 14: Applications IV (Hall 17 - chairman: M. Tanaka)

16.45 P. Arena, S. Baglio, L. Fortuna, G. Manganaro
CNN Processing for NMR Spectra
17.00 P. Arena, L. Fortuna, G. Manganaro, S. Spina
CNN Image Processing for the Automatic Classification of Oranges
17.15 S. Schwarz
Detection of Defects on Photolitographic Masks by Cellular Neural 
Networks
17.30 M. Ikegami, M. Tanaka
Moving Image Coding and Decoding by DTCNN with 3-D Templates
17.45 M. Kanaya, M. Tanaka
Robot Multi-Driving Controls by Cellular Neural Networks

GENERAL INFORMATION

Venue: All sessions will be held at the Faculty of Engineering, "La 
Sapienza" University of Rome, vie Eudossiana, 18 - 00184 Rome

Coffee Breaks and lunches are included in the registration fee; they will 
be served in the room of the Frescoes, opposite the main lecture hall. In 
order to receive them, it is necessary to wear the workshop badge.

Banquet will be held at restaurant "4 Colonne", via della Posta Vecchia, 
4, near Piazza Navona, tel. 68307152/68805261. It can be reached on foot 
in about 20 minutes from the Faculty of Engineering, by bus no. 81 or 87 
from via dei Fori Imperiali to Corso Rinascimento, or by taxi. You may 
ask for directions at the registration desk. Please bring the coupon you 
receive at registration.

Communications: You may receive messages by fax, no. +39-6-4742647. Faxes 
should be addressed to your name, with clear statement "participant to 
CNNA-94" in the cover sheet. A UNIX ASCII terminal will be available for 
all participants, with full Internet capabilities. You may receive 
messages at the usual e-mail address of the workshop: 
cnna at tce.ing.uniroma1.it  Subject of messages should include your name.

Public telephones, suitable for local, long-distance, and international 
calls, are available in the Faculty. They accept magnetic cards that can 
be purchased at the registration desk, at tobacconists' and newsstands, 
and in other shops carrying the Telecom symbol.

Bank: Banca di Roma has an agency within the Faculty of Engineering. It 
is open in the morning from 8.25 to 1.35. Several other banks are 
available in the area of the University, also open from 3 to 4pm. Some 
cash dispensers, e.g. the one at Banca di Roma on via Cavour, between 
Hotel Palatino and via dei Fori Imperiali, accept credit cards for cash 
retrieval 24 hours a day. 

Authors are requested to meet with their session chairman 15 minutes 
prior to the session in the same room where it will be held. Time slots 
for regular papers are strictly limited to 15 minutes, including 
discussion. Poster authors should set up posters on 10.30, Wed. Dec. 21, 
and remove them during lunch break.

Demonstrations of working hardware and software for CNNs will be held 
throughout the workshop during breaks. Demonstration authors should bring 
their material for set up on 8am of the day agreed upon, and remove it at 
the end of sessions.


FURTHER INFORMATION
Please contact:
CNNA-94
tel. +39-6-44585836
fax. +39-6-4742647
e-mail cnna at tce.ing.uniroma1.it


From ai at aaai.org  Mon Dec 12 15:36:08 1994
From: ai at aaai.org (AAAI)
Date: Mon, 12 Dec 94 12:36:08 PST
Subject: AAAI 1995 Fall Symposium Series Call for Participation
Message-ID: <9412122036.AA19948@aaai.org>

AAAI 1995 Fall Symposium Series 
Call for Participation

November 10-12, 1995

Massachusetts Institute of Technology 
Cambridge, Massachusetts

Sponsored by the American Association for Artificial Intelligence 
445 Burgess
Drive, Menlo Park, CA 94025 
(415) 328-3123 (voice)
(415) 321-4457 (fax)
fss at aaai.org

The American Association for Artificial Intelligence presents the 1995 Fall
Symposium Series, to be held Friday through Sunday, November 10-12, 1995, at 
the Massachusetts Institute of Technology.

The topics of the eight symposia in the 1995 Fall Symposium Series are:

-  Active Learning 
-  Adaptation of Knowledge for Reuse 
-  AI Applications in Knowledge Navigation and Retrieval 
-  Computational Models for Integrating Language and Vision 
-  Embodied Language and Action 
-  Formalizing Context 
-  Genetic Programming 
-  Rational Agency: Concepts, Theories, Models, and Applications

Symposia will be limited to between forty and sixty participants. Each
participant will be expected to attend a single symposium. Working notes 
will be prepared and distributed to participants in each symposium.

A general plenary session, in which the highlights of each symposium will be
presented, will be held on Saturday, November 11, and an informal reception 
will be held on Friday, November 10.

In addition to invited participants, a limited number of other interested
parties will be able to register in each symposium on a first-come, 
first-served basis. Registration will be available by 1 August, 1995. 
To obtain registration information write to the AAAI at 445 Burgess Drive, 
Menlo Park, CA 94025 (fss at aaai.org).

Submission Dates

-  Submissions for the symposia are due on April 14, 1995. 
-  Notification of acceptance will be given by May 19, 1995. 
-  Material to be included in the working notes of the symposium must be
   received by September 1, 1995.

See the appropriate section below for specific submission requirements for each
symposium.

This document is available as 
http://www.ai.mit.edu/people/las/aaai/fss-95/fss-95-cfp.html

*******************************************************************************

			ACTIVE LEARNING

An active learning system is one that can influence the training data it
receives by actions or queries to its environment. Properly selected, these
actions can drastically reduce the amount of data and computation required by a
machine learner. Active learning has been studied independently by researchers
in machine learning, neural networks, robotics, computational learning theory,
experiment design, information retrieval, and reinforcement learning, among
other areas. This symposium will bring researchers together to clarify the
foundations of active learning and point out synergies to build on.

Submission Information

Potential participants should submit a position paper (at most two pages)
discussing what the participant could contribute to a dialogue on active
learning and/or what they hope to learn by participating. Suggested topics
include:

Theory: What are the important results in the theory of active learning and what
are important open problems? How much guidance does theory give to application?

Algorithms: What successful algorithms have been found for active learning? How
general are they? For what tasks are they appropriate?

Evaluation: How can accuracy, convergence, and other properties of active
learning algorithms be evaluated when, for instance, data is not sampled
randomly?

Taxonomy: What kinds of information are available to learners (e.g. membership
vs. equivalence queries, labeled vs. unlabeled data) and what are the ways
learning methods can use them? What are the commonalities among methods studied
by different fields?

Papers should be sent to David D. Lewis, lewis at research.att.com, AT&T Bell
Laboratories, 600 Mountain Ave., Room 2C-408, Murray Hill, NJ 07974-0636.
Electronic mail submissions are strongly preferred.

Symposium Structure

The symposium will be broken into sessions, each dedicated to a major theme
identified within the position papers. Sessions will begin with a background
presentation by an invited speaker, followed by brief position statements from
selected participants. A significant portion of each session will be reserved
for group discussion, guided by a moderator and focused on the core issue for
the session. The final session of the symposium will accommodate new issues that
are raised during sessions.

Organizing Committee

David A. Cohn (cochair), MIT, cohn at psyche.mit.edu; David D. Lewis (cochair),
AT&T Bell Labs, lewis at research.att.com; Kathryn Chaloner, U. Minnesota ; Leslie
Pack Kaelbling, Brown U.; Robert Schapire, AT&T Bell Labs; Sebastian Thrun, U.
Bonn; Paul Utgoff, U. Mass Amherst.

******************************************************************************

		ADAPTATION OF KNOWLEDGE FOR REUSE

Several areas in AI address issues of creating and storing knowledge constructs
(such as cases, plans, designs, specifications, concepts, domain theories,
schedules). There is broad interest in reusing these constructs in similar
problem-solving situations so as to avoid expensive re-derivation. Adaptation
techniques have been developed to support reuse in frameworks such as analogical
problem solving, case-based reasoning, problem reformulation, or representation
change and task domains such as creativity, design, planning, program
transformation or software reuse, schedule revision, and theory revision.

However, many open issues remain, and progress on such issues as case adaptation
would substantially assist many researchers and practitioners. Our goals are to
characterize the approaches to adaptation employed in various AI subfields,
define the core issues in adaptation of knowledge, and advance the
state-of-the-art in addressing these issues. We intend that presentations will
investigate novel solutions to unsolved problems on adaptation, reflect diverse
viewpoints, and focus on adaptation issues that are common to several subfields
of AI. Discussions will be held on the strengths and limitations of adaptation
techniques and their interrelationships.

Invited talks will be given by experts who will discuss methods for the
adaptation of various types of knowledge constructs. Two panels will be held.
First, researchers studying knowledge adaptation from different perspectives
will discuss how approaches used in their community differ from those used
elsewhere, focusing on their potential benefits for other problems. Panelists in
the second panel will lead discussions on identifying the core issues in
knowledge adaptation raised in the presentations and the impact of the proposed
methods on addressing these issues.

Submission Information

Anyone interested in presenting relevant material is invited to email PostScript
submissions to aha at aic.nrl.navy.mil using the six-page AAAI-94 proceedings
format. Anyone interested in attending is asked to submit a two-page research
statement and a list of relevant publications. Please see
http://www.aic.nrl.navy.mil/~aha/aaai95-fss/home.html for further information.

Organizing Committee

David W. Aha (cochair), NRL, aha at aic. nrl.navy.mil; Brian Falkenhainer, Xerox;
Eric K. Jones, Victoria University; Subbarao Kambhampati, Arizona State
University; David Leake, Indiana University; Ashwin Ram (cochair), Georgia
Institute of Technology, ashwin at cc.gatech.edu.

*****************************************************************************

	AI APPLICATIONS IN KNOWLEDGE NAVIGATION AND RETRIEVAL

The diversity and volume of accessible on-line data is increasing dramatically.
As a result, existing tools for searching and browsing information are becoming
less effective. The increasing use of non-text data such as images, audio and
video has amplified this trend.

Knowledge navigation systems are knowledge-based interfaces to information
resources. They allow users to investigate the contents of complex and diverse
sources of data in a natural manner. For example, intelligent browsers that can
help direct a user through a large multi-dimensional information space, agents
that users can direct to perform information finding tasks, or knowledge-based
intermediaries that employ retrieval strategies to gather information relevant
to a userUs request.

The purpose of this symposium is to examine the state of the art in knowledge
navigation by examining existing applications and by discussing new techniques
and research directions. We encourage two types of submissions: work-in-progress
papers that point towards the future of this research area, and demonstrations
of knowledge navigation systems. Some research issues of interest:

-  Indexing: What indexing methods are appropriate and feasible for knowledge
   navigation systems? How can indices be extracted from data?

-  Retrieval: What retrieval methods are appropriate for knowledge navigation?
   What retrieval strategies can be employed?

-  Learning: How can knowledge navigation systems adapt to a changing knowledge
   environment and to user needs?

-  User interfaces: What are the characteristics of a useful navigational
   interface? What roles can or should an "agent" metaphor play in such 
   interfaces? How can a navigation system orient the user in the 
   information space?

-  Multi-source integration: How can multiple data and knowledge sources be
   integrated to address usersU needs?

-  Multimedia: What are the challenges presented by multimedia information
   sources?

Submission Information

The symposium will consist of invited talks, presentations, and hands-on
demonstration/discussion sessions. Interested participants should submit a short
paper (8 pages maximum) addressing a research issue in knowledge navigation or
describing a knowledge navigation system that can be made available for hands-on
demonstration at the symposium. System descriptions should clearly indicate the
novel and interesting features of the system to be presented and its
applicability to the central problems in knowledge navigation. Those wishing to
demonstrate should also include a one-page description of their hardware and
connectivity requirements. Send, by email,  either a URL pointing to a
PostScript version of the paper or the PostScript copy itself to 
aiakn at cs.uchicago.edu. Or, send 5 hard copies to Robin Burke, AI Applications 
in Knowledge Navigation, University of Chicago, Department of Computer Science.
1100 E. 58th St. Chicago, IL 60637. For further information, a web page for this
symposium is located at http://www-cs.uchicago.edu/ ~burke/aiakn.html

Organizing Committee

Robin Burke (chair), University of Chicago, burke at cs.uchicago.edu; Catharine
Baudin, NASA Ames; Su-Shing Chen, National Science Foundation; Kristian Hammond,
University of Chicago; Christopher Owens, Bolt, Beranek & Newman.

Program Committee

Ray Bariess, Institute for the Learning Sciences; Alon Levy, AT&T Bell
Laboratories; Jim Mayfield, University of Maryland, Baltimore County; Dick
Osgood, Andersen Consulting.

******************************************************************************

	COMPUTATIONAL MODELS FOR INTEGRATING LANGUAGE AND VISION

This symposium will focus on research issues in developing computational models
for integrating language and vision. The intrinsic difficulty of both natural
language processing and computer vision has discouraged researchers from
attempting integration, although in some cases it may simplify individual tasks
like collateral-based vision, resolving ambiguous sentences through the use of
visual information.

Developing a bridge between language and vision is nontrivial, because the
correspondence between words and images is not one-to-one. Much has been said
about the necessity of linking language and perception for a system to exhibit
intelligent behavior, but there has been relatively little work on developing
computational models for this task. A natural-language understanding system
should be able to understand and make references to the visual world. The use of
scene-specific context (obtained from written or spoken text accompanying a
scene) could greatly enhance the performance of computer vision systems.

Some topics to be addressed are:

-  use of collateral text in image and graphics understanding

-  generating natural-language descriptions of visual data (e.g., event
   perception in image sequences)

-  identifying and extracting visual information from language

-  understanding spatial language, spatial reasoning

-  knowledge representation for linguistic and visual information, hybrid
   (language and visual) knowledge bases

-  use of visual data in disambiguating/understanding text

-  content-based retrieval from integrated text/image databases

-  language-based scene modeling (e.g., picture or graphics generation)

-  cognitive theories connecting language and perception

Submission Information

The symposium will consist of invited talks, panel discussions, individual
presentations and group discussions. Those interested in making a presentation
should submit a technical paper (not to exceed 3,000 words). Other participants
should submit either a position paper or a research abstract. Email submissions
in postscript format are encouraged. Send to burhans @cs.buffalo.edu.
Alternatively, 4 hard copies may be sent to Rohini Srihari, CEDAR/SUNY at
Buffalo, UB Commons, 520 Lee Entrance Suite 202, Buffalo, NY 14228-2567. Further
information on this symposium may be found at
http://www.cedar.buffalo.edu/Piction/FSS95/CFP.html. Please address questions to
Debra Burhans (burhans at cs.buffalo.edu) or Rajiv Chopra (rchopra at cs.buffalo.edu).

Organizing Committee

Janice Glasgow, Queen's University; Ken Forbus, Northwestern University; Annette
Herskovits, Wellesley College; Gordon Novak, University of Texas at Austin;
Candace Sidner, Lotus Development Corporation; Jeffrey Siskind, University of
Toronto; Rohini K. Srihari (chair), CEDAR, SUNY at Buffalo,
rohini at cedar.buffalo. edu; Thomas M. Strat, SRI International; David Waltz, NEC
Research Institute.

*******************************************************************************

			EMBODIED LANGUAGE AND ACTION

This symposium focuses on agents that can use language or similar communication,
such as gesture, to facilitate extended interactions in a shared physical or
simulated world. We examine how this embodiment in a shared world both
stimulates communication and provides a resource for understanding it. Our focus
is on the design of artificial agents, implemented in software, hardware, or as
animated characters. Papers should clearly relate the technical content
presented to one of the following tasks:

-  Two or more communicating agents work together to construct, carry out
   maintenance on, or destroy a physical or simulated artifact (Collaborative
   Engagement)

-  An agent assists a human by fetching or delivering physical or software
   objects. The human communicates with the agent about what is to be fetched or
   delivered to where. (Delivery Assistance)

We solicit papers on the following issues (not to the exclusion of others):

-  Can task contexts act as resources for communication by simplifying the
   interpretation and production of communicative acts?

-  How does physical embodiment and its concomitant resource limitation affect
   an agentUs ability to interpret or generate language?

-  Can architectures designed to support perception and action support language
   or other forms of communication?

-  How can agents to mediate between the propositional representations of
   language and the (often) non-propositional representations of perception and
   action?

-  What tradeoffs exist between the use of communication to improve the agents'
   task performance and the additional overhead involved in understanding and
   generating messages?

-  Do differences between communication used to support concurrent task
   execution and communication used to support planning, reflect deeper 
   differences in agent ability?

-  What is the role of negotiation, whether of task responsibilities, or of
   reference and meaning, in such situated task environments?

Submission Information

Interested participants should submit either (1) a paper (in 12 pt font, not to
exceed 3000 words), or (2) a statement of interest briefly describing the
authorUs relevant work in this area and listing recent relevant publications.
Send contributions, plain ascii or postscript, to ian at ai. mit.edu. If electronic
submission is impossible, mail 6 copies to Ian Horswill, MIT Artificial
Intelligence Laboratory, 545 Technology Square, Cambridge, MA 02139.

Organizing Committee

John Batali, UCSD; Jim Firby, University of Chicago; Ian Horswill (cochair),
MIT, ian at ai.mit.edu; Marilyn Walker (cochair), Mitsubishi Cambridge Research
Labs, walker at merl.com; Bonnie Webber, University of Pennsylvania.

******************************************************************************

			FORMALIZING CONTEXT

The notion of context has played an important role in AI systems for many years.
However, formal logical explication of contexts remains an area of research with
significant open issues. This symposium will provide a forum for discussing
formalizations of contexts, approaches to resolving open issues, and application
areas for context formalisms.

The most ambitious goal of formalizing contexts is to make automated reasoning
systems which are never permanently stuck with the concepts they use at a given
time because they can always transcend the context they are in. Such a
capability would allow the designer of a reasoning system to include only such
phenomena as are required for the system's immediate purpose, retaining the
assurance that if a broader system is required later, "lifting rules" can be
devised to restate the facts from the narrow context in the broader context with
qualifications added as necessary. A formal theory of context in which sentences
are always considered as asserted within a context could provide a basis for
such transcendence.

Formal theories of context are also needed to provide a representation of the
context associated with a particular circumstance, e.g. the context of a
conversation in which terms have particular meanings that they wouldn't have in
the language in general. Linguists and philosophers have already studied similar
notions of context. An example is the situation theory that has been proposed in
philosophy and applied to linguistics. However, these theories usually lie
embedded in the analysis of specific linguistic constructions, so locating the
exact match with AI concerns is itself a research challenge.

This symposium aims to bring together researchers who have studied or applied
contexts in AI or related fields. Technical papers dealing with formalizations
of context, the problem of generality, and use of context in common sense
reasoning are especially welcome. However, survey papers which focus on contexts
from other points of view, such as philosophy, linguistics, or natural language
processing, or which apply contexts in other areas of AI, are also encouraged.

Submission Information

Persons wishing to make presentations should submit papers (up to 12 pages, 12
pt font). Persons wishing only to attend should submit a 1-2 page research
summary including a list of relevant publications. A PostScript file or 8 paper
copies should be sent to the program chair, Sasa Buvac Department of Computer
Science Stanford University Stanford CA 94305-2140 buvac at sail.stanford.edu.
Limited funding will be available to support student travel.

Organizing Committee

Sasa Buvac (chair), Stanford University, buvac at sail.stanford.edu; Richard Fikes,
Stanford University; Ramanathan Guha, MCC; Pat Hayes, Beckman Institute; John
McCarthy, Stanford University; Murray Shanahan, Imperial College; Robert
Stalnaker, MIT; Johan van Benthem, University of Amsterdam. Genetic programming
(GP) extends the genetic algorithm to the domain of computer programs. In
genetic programming, populations of programs are genetically bred to solve
problems. Genetic programming can solve problems of system identification,
classification, control, robotics, optimization, game-playing, and pattern
recognition.

******************************************************************************

			GENETIC PROGRAMMING

Starting with a primordial ooze of hundreds or thousands of randomly created
programs composed of functions and terminals appropriate to the problem, the
population is progressively evolved over a series of generations by applying the
operations of Darwinian fitness proportionate reproduction and crossover (sexual
recombination).   
Topics of interest for the symposium include:
 
-  The theoretical basis of genetic programming 
-  Applications of genetic programming
-  Rigorousness of validation techniques 
-  Hierarchical decomposition, e.g. automatically defined functions 
-  Competitive coevolution 
-  Automatic parameter tuning 
-  Representation issues 
-  Genetic operators 
-  Establishing standard benchmark problems 
-  Parallelization techniques 
-  Innovative variations 

Submission Information

The format of the symposium will encourage interaction and discussion, but 
will also include formal presentations. Persons wishing to make a presentation 
should submit an extended abstract of up to 2500 words of their work in 
progress or completed work. For accepted abstracts, full papers will be due 
at a date closer to the symposium.

Persons not wishing to make a presentation are asked to submit a one-page 
description of their research interests since there may be limited room 
for participation.

Submit your abstract or one-page description as plain text electronically 
by Friday April 14, 1995, with a hard-copy backup to Eric V. Siegel, AAAI 
GP Symposium Co-Chair, Columbia University, Department of Computer Science,
500 W 120th Street, New York, NY 10027, USA; telephone: 212-939-7112, 
fax: 212-666-0140, e-mail: evs at cs.columbia.edu.

Organizing Committee

Robert Collins, USAnimation, Inc.; Frederic Gruau, Stanford University; 
John R. Koza (co-chair), Stanford University, koza at cs.stanford.edu; 
Robert Collins, US Animation, Inc.; Conor Ryan, University College Cork; 
Eric V. Siegel (co-chair), Columbia University, evs at cs.columbia.edu; 
Andy Singleton, Creation Mechanics, Inc.; Astro Teller, Carnegie-Mellon 
University.

******************************************************************************

RATIONAL AGENCY: CONCEPTS, THEORIES, MODELS, AND APPLICATIONS

This symposium explores conceptions of rational agency and their implications
for theory, research, and practice. The view that intelligent systems are, or
ought to be, rational agents underlies much of the theory and research in
artificial intelligence and cognitive science. However, no consensus exists on a
proper view of agency or rationality principles for practical agents.

Traditionally agents are presumed disposed toward purposive action. However,
agent theories abound in which behavior is fundamentally reactive. Some theories
emphasize agents' abilities to manage private belief systems. Others focus on
agents' interactions with their environment, sometimes including other agents.
Application builders have recently broadened the term "agent" to mean any
embedded system performing tasks to support human users.

Rationality accounts are equally diverse. Rationality involves having reasons
warranting particular beliefs (epistemic rationality) or particular desires and
actions (strategic or practical rationality). Many agent models propose
epistemic rationality criteria such as logical consistency or consequential
closure. Other agent models base practical rationality on classical or
non-monotonic logics for reasoning about action. Such logicist views are now
being challenged by decision theoretic accounts emphasizing optimal action under
uncertainty, including recent work on decision theoretic principles of
metareasoning for limited rationality.

Our symposium will explore the diverse views of rational agency. Through
informal presentations and group discussion, participants will critically
examine agency concepts and rationality principles, review computational agent
models and applications, and consider promising directions for future work on
this topic.

Submission Information

Prospective participants should submit a brief paper (5 pages or less)
describing their research in relation to any of the following questions:

-  Is rationality important; must an agent be rational to be successful?

-  What are suitable principles of epistemic, strategic, and limited
   rationality?

-  Are rationality principles applicable to retrospective processes such as
   learning?

-  What are general requirements on rational agent architectures?

-  How, if at all, must a model of rational agency be modified to account for
   social, multi-agent interaction?

Those wishing to make a specific presentation should describe its contents in
their concept paper. Note: While we recognize that our topic lends itself to
formal analysis, we also encourage discussion of experimental work with
implemented agents. Postscript files of concept papers should be sent by email
only to the program chair, fehling at lis.stanford.edu.

Organizing Committee

Michael Fehling (chair), Stanford University, fehling at lis.stanford.edu; Don
Perlis, University of Maryland; Martha Pollack, University of Pittsburgh; John
Pollock, University of Arizona.






From tesauro at watson.ibm.com  Thu Dec  8 14:28:41 1994
From: tesauro at watson.ibm.com (tesauro@watson.ibm.com)
Date: Thu, 8 Dec 94 14:28:41 EST
Subject: NIPS*94 Top Ten List
Message-ID: <mailman.715.1149591321.29955.connectionists@cs.cmu.edu>

By popular demand, here once again is the Top Ten List.
Thanks go to my comedy-writing collaborators, Mike Mozer
and Don Mathis of the Univ. of Colorado.
--Gerry Tesauro
------------------------------------------------------------------
         Top 10 Little-Known Improvements to NIPS This Year

10.  No more of that annoying math!

 9.  To keep speakers on their toes, a gong at every table.

 8.  Special prize for "Most Laughable Poster"

 7.  Push pins available to everyone, not just to poster presenters.

 6.  Special evening session in which senior researchers
     display their surgical scars.

 5.  Hotel staff all licensed to back-propagate.

 4.  More rock, less talk.

 3.  Student volunteers at the registration desk will finally
     stop asking "You want fries with that?"

 2.  Invited speakers disqualified if they test positive
     for steroids.

 1.  Morgan Kaufmann to publish special "Swimsuit Edition"
     of conference proceedings.


From dwang at cis.ohio-state.edu  Mon Dec 12 10:50:23 1994
From: dwang at cis.ohio-state.edu (DeLiang Wang)
Date: Mon, 12 Dec 1994 10:50:23 -0500
Subject: About sequential learning (or interference)
Message-ID: <mailman.716.1149591321.29955.connectionists@cs.cmu.edu>

It was reported some time ago that multilayer perceptrons suffer the problem
of so-called "catastrophic interference", meaning that later training will
destroy previously acquired "knowledge" (see McCloskey & Cohen, Psychol. of
Learning and Motivat. 24, 1990; Ratcliff, Psychol. Rev. 97, 1990). This seems
to be a serious problem, if we want to use neural networks both as a stable
knowledge store and a long-term problem solver.

The problems seems to exist in associative memory models as well, even though
the simple prescription of the Hopfield net can easily incorporate more
patterns as long as they are within the capacity limit. But we all know that
the original Hopfield net does not work well for correlated patterns, which
are the most interesting ones for real applications. Kanter and Sompolinsky
proposed a prescription to tackle the problem (Phys. Rev. A, 1987). 
Their prescription, however, requires nonlocal learning. 

Diederich and Opper (1987, Phys. Rev. Lett.) later proposed a local, 
iterative learning rule (similar to perceptron learning) that they show will 
converge to the prescription (which is the old idea of orthogonalization). 
According to their paper, to learn a new pattern one needs to bring in all 
previously acquired patterns during iterative training in order 
to make all of the patterns converge to the desired prescription.  Because
of this learning scheme, I suspect that the algorithm of Diederich and Opper
also suffers "catastrophic forgetting".

Is that a fair assessment of the problem?  Has any major effort been taken to 
address this important problem? 

DeLiang Wang


From jaap.murre at mrc-apu.cam.ac.uk  Tue Dec 13 11:59:33 1994
From: jaap.murre at mrc-apu.cam.ac.uk (Jaap Murre)
Date: Tue, 13 Dec 94 16:59:33 GMT
Subject: hypertransfer and interference
Message-ID: <9412131659.AA03848@rigel.mrc-apu.cam.ac.uk>

DeLiang Wang asks whether any major effort has been undertaken to
investigate and possibly remedy the problem of catastrophic interference.

A fairly detailed analysis of this problem can be found in:

Murre, J.M.J. (1992b). Categorization and learning in modular neural
      networks. Hemel Hempstead: Harvester Wheatsheaf. Co-published
      by Lawrence Erlbaum in the USA and Canada (Hillsdale, NJ).

I have recently completed a paper that shows that backpropagation not
only suffers from 'catastrophic interference' but also from 'hyper
transfer', i.e., in some circumstances performance on a set A actually
*improves* when learning a second set B. The learning transfer effects
are catastrophic (or hyper) with respect to human learning data. The
paper also shows that two-layer networks do not suffer from excessive
transfer and are in fact in very close accordance with the human
interference data as summarized in the classic paper by Osgood (1949):

Murre, J.M.J. (in press). Transfer of learning in backpropagation
      networks and in related neural network models. To appear in
      Levy, Bairaktaris, Bullinaria, & Cairns (Eds.), Connectionist
      Models of Memory and Language. London: UCL Press. (in fpt
      directory: hyper1.ps)

I have put the paper in our anonymous ftp directory:

      ftp ftp.mrc-apu.cam.ac.uk
      cd /pub/nn/murre
      bin
      get hyper1.ps



From lpratt at franklinite.Mines.Colorado.EDU  Tue Dec 13 05:20:35 1994
From: lpratt at franklinite.Mines.Colorado.EDU (Lorien Y. Pratt)
Date: Tue, 13 Dec 94 11:20:35 +0100
Subject: About sequential learning (or interference)
Message-ID: <9412131020.AA01764@franklinite.Mines.Colorado.EDU>

Deliang,

My own work on transfer between neural networks has addressed the sequential 
learning problem, see pratt-93, cited below, and lots of other papers available 
from my web page: http://vita.mines.colorado.edu:3857/1s/lpratt.  My formuation 
differs from others in that, if the new task is essentially different from the 
old task, I intentionally lose old-task performance if necessary to improve on 
the new task.  Other people who have looked at a similar formulation include 
sharkey-92, sharkey-93, naik-92b, and agarwal-92, martin-88.  Those who have 
attempted to preserve old-task performance include mccloskey-89, and recent 
work by Thrun & Mitchell, especially thrun-93a, but also including thrun-92, 
thrun-93b. Sebastian and I had a long talk at NIPS about a general formulation 
for transfer as a solution to the sequential learning problem -- he's got an 
excellent formulation along these lines.

Approaches to modular training can also be viewed as one approach to handling 
the sequential learning problem, in the special case where the old tasks can be 
viewed as subtasks of the new task (i.e. pratt-91, waibel-89, and Jacobs' and 
others' recent work on mixtures of experts, including Pomerleau's approach to 
Alvinn, see last year's NIPS).

My thesis (see my web page) has a more detailed review of all this stuff.

Hope that helps.
  --Lori, recovering from NIPS
  

  @incollection{ pratt-93,
MYKEY           = " pratt-93 : .con .bap",
EDITOR          = "C.L. Giles and S. J. Hanson and J. D. Cowan",
BOOKTITLE       = "{Advances in Neural Information Processing Systems 5}",
AUTHOR          = "L. Y. Pratt",
TITLE           = "Discriminability-Based Transfer between Neural Networks",
ADDRESS         = "San Mateo, CA",
PUBLISHER       = "Morgan Kaufmann Publishers",
YEAR            = 1993,
PAGES           = {204--211},
CATALOGDATE     = "April 12, 1993",
NOTE            = "Also available via anonymous ftp to
franklinite.mines.colorado.edu: pub/pratt-papers/pratt-nips5.ps.Z",
}
@inproceedings{ sharkey-92,
MYKEY           = " sharkey-92 : ",
TITLE           = "Adaptive generalisation and the transfer of knowledge",
AUTHOR          = "Noel E. Sharkey and Amanda J. C. Sharkey",
BOOKTITLE       = "Proceedings of the Second Irish Neural Networks Conference,
                   Belfast",
YEAR            = 1992,
CATALOGDATE     = "December 18, 1992",
}

@article{ sharkey-93,
MYKEY           = " sharkey-93 : ",
TITLE           = "Adaptive generalisation and the transfer of knowledge",
AUTHOR          = "N. E. Sharkey and A. J. C. Sharkey",
NUMBER          = "In press",
VOLUME          = "Special Issue on Connectionism",
JOURNAL         = "AI Review",
YEAR            = 1993,
ANNOTE          = "In press",
CATALOGDATE     = "May 19, 1993",
}
@inproceedings{ naik-92b,
MYKEY           = " naik-92b : .eml .unb .con ",
AUTHOR          = "D. K. Naik and R. J. Mammone and A. Agarwal",
TITLE           = "Meta-Neural Network approach to learning by learning",
YEAR            = 1992,
BOOKTITLE       = "Intelligence Engineering Systems through Artificial Neural
                   Networks",
ORGANIZATION    = "The American Society of Mechanical Engineers",
PUBLISHER       = "ASME Press",
VOLUME          = 2,
PAGES           = {245--252},
CATALOGDATE     = "December 23, 1992",
}

@inproceedings{ agarwal-92,
MYKEY           = " agarwal-92 : .eml .unb .con ",
AUTHOR          = "A. Agarwal and R. J. Mammone and D. K. Naik",
TITLE           = "An on-line Training Algorithm to Overcome Catastrophic
                   Forgetting",
BOOKTITLE       = "Intelligence Engineering Systems through Artificial Neural
                   Networks",
YEAR            = 1992,
ORGANIZATION    = "The American Society of Mechanical Engineers",
PUBLISHER       = "ASME Press",
VOLUME          = 2,
PAGES           = {239--244},
CATALOGDATE     = "December 23, 1992",
}

@techreport{ martin-88,
MYKEY           = " martin-88 : ",
TITLE           = "The Effects of Old Learning on New in Hopfield and
                   Backpropagation Nets",
AUTHOR          = "Gale Martin",
KEY             = "martin-88",
INSTITUTION     = "Microelectronics and Computer Technology Corporation (MCC)",
NUMBER          = "ACA-HI-019",
YEAR            = 1988,
CATALOGDATE     = "January 3, 1992",
}

@article{ mccloskey-89,
MYKEY           = " mccloskey-89 : .unb .con .csy .adap",
TITLE           = "Catastrophic interference in connectionist networks: the
                   sequential learning problem",
KEY             = "mccloskey-89",
AUTHOR          = "Michael McCloskey and Neal J. Cohen",
JOURNAL         = "The psychology of learning and motivation",
VOLUME          = 24,
YEAR            = 1989,
CATALOGDATE     = "April 9, 1991",
}
@inproceedings{ thrun-93a,
MYKEY           = " thrun-93a : ",
TITLE           = "Lifelong Robot Learning",
AUTHOR          = "Sebastian B. Thrun and Tom M. Mitchell",
BOOKTITLE       = "Proceedings of the {NATO} {ASI}: The biology and technology
                   of intelligent autonomous agents",
EDITOR          = "Luc Steels",
YEAR            = 1993,
KEY             = "thrun-93a",
CATALOGDATE     = "September 1, 1993",
}

@INPROCEEDINGS{thrun-93b,
AUTHOR        = {Thrun, Sebastian B. and Mitchell, Tom M.},
TITLE         = {Integrating Inductive Neural Network Learning and
                 Explanation-Based Learning},
BOOKTITLE     = {Proceedings of IJCAI-93},
YEAR          = {1993},
ORGANIZATION  = {IJCAI, Inc.},
PUBLISHER     = {},
ADDRESS       = {Chamberry, France},
MONTH         = {},
CATALOGDATE     = "October 11, 1993",
}

@inproceedings{ pratt-91,
MYKEY           = " pratt-91 : .min .bap .app .spc .con ",
AUTHOR          = "Lorien Y. Pratt and Jack Mostow and Candace A. Kamm",
TITLE           = "Direct Transfer of Learned Information among Neural
                    Networks",
BOOKTITLE       = "Proceedings of the Ninth National Conference on
                   Artificial Intelligence (AAAI-91)",
PAGES           = {584--589},
ADDRESS         = "Anaheim, CA",
YEAR            = 1991,
}

@article{ waibel-89,
MYKEY           = " waibel-89 : .bap .unr .unb .tem .spc .con ",
TITLE           = "Modular Construction of Time-Delay Neural Networks for
                    Speech Recognition",
AUTHOR          = "Alexander Waibel",
journal         = "Neural Computation",
volume          = 1,
pages           = {39--46},
year            = 1989
}



From wermter at nats2.informatik.uni-hamburg.de  Tue Dec 13 06:04:05 1994
From: wermter at nats2.informatik.uni-hamburg.de (Stefan Wermter)
Date: Tue, 13 Dec 94 12:04:05 +0100
Subject: book on hybrid connectionist language processing
Message-ID: <9412131104.AA01259@nats2.informatik.uni-hamburg.de>



BOOK ANNOUNCEMENT
-----------------

The following book is now available from the beginning of December 1994.





Title:      Hybrid connectionist natural language processing

Date:       1995

Author:     Stefan Wermter
            Dept. of Computer Science
            University of Hamburg
            Vogt-Koelln-Str. 30
            D-22527 Hamburg
            Germany

            wermter at informatik.uni-hamburg.de

Series:     Neural Computing Series 7

Publisher:  Chapman & Hall Inc
            2-6 Boundary Row
            London SE1 8HN
            England



(Order information in the end of this message)



Description
-----------

The objective of this  book is  to describe a  new approach  in hybrid
connectionist natural language processing which bridges the gap between
strictly symbolic and connectionist systems. This objective is tackled
in two ways: the book gives an overview of hybrid connectionist archi-
tectures for natural  language processing; and  it demonstrates that a
hybrid connectionist architecture can be used for learning real-world
natural language problems. The book is primarily intended for scientists
and students interested in the fields of artificial intelligence, neural
networks,  connectionism, natural  language processing, hybrid symbolic
connectionist architectures,  parallel distributed processing, machine
learning, automatic knowledge  acquisition or computational linguistics.
Furthermore,  it might  be of  interest for  scientists and students
in information retrieval and cognitive science, since the book points
out interdisciplinary relationships to these fields.

We develop a systematic spectrum of hybrid connectionist architectures,
from completely symbolic architectures to separated hybrid connectionist
architectures,  integrated  hybrid  connectionist architectures and
completely connectionist architectures. Within this systematic spectrum
we have designed a system SCAN with two separated hybrid  connectionist
architectures and two integrated hybrid connectionist architectures for
a scanning  understanding of phrases. A scanning understanding  is a
relation-based flat understanding in  contrast to traditional symbolic
in-depth understanding. Hybrid  connectionist representations  consist
of either a combination of connectionist and symbolic representations
or different  connectionist representations. In particular, we focus on
important  tasks like structural  disambiguation and semantic context
classification. We show that  a parallel modular, constraint-based,
plausibility-based  and learned use of multiple hybrid connectionist
representations provides powerful architectures  for learning a scanning
understanding.  In particular, the combination of direct encoding of
domain-independent structural knowledge and the connectionist learning of
domain-dependent semantic knowledge, as suggested by a scanning under-
standing in SCAN, provides concepts which lead to  flexible, adaptable,
transportable architectures for different domains.




Table of Contents
-----------------

1 Introduction
  1.1 Learning a Scanning Understanding
  1.2 The General Approach
  1.3 Towards a Hybrid Connectionist Memory Organization
  1.4 An Overview of the SCAN Architecture
  1.5 Organization and Reader's Guide

2 Connectionist and Hybrid Models for Language Understanding
  2.1 Foundations of Connectionist and Hybrid Connectionist Approaches
  2.2 Connectionist Architectures
    2.2.1 Representation of Language in Parallel Spatial Models
      Early Pattern Associator for Past Tense Learning
      Pattern Associator for Semantic Case Assignment
      Pattern Associator with Sliding Window
      Time Delay Neural Networks
    2.2.2 Representation of Language in Recurrent Models
      Recurrent Jordan Network for Action Generation
      Simple Recurrent Network for Sequence Processing
      Recursive Autoassociative Memory Network
    2.2.3 Towards Modular and Integrated Connectionist Models
      Cascaded Networks
      Sentence Gestalt Model
      Grounding Models
  2.3 Hybrid Connectionist Architectures
    2.3.1 Sentence Analysis in Hybrid Models
      Hybrid Interactive Model for Constraint Integration
      Hybrid Model for Sentence Analysis
    2.3.2 Inferencing in Hybrid Models
      Symbolic Marker Passing and Localist Networks
      Symbolic Reasoning with Connectionist Models
    2.3.3 Architectural Issues in Hybrid Connectionist Systems
      Symbolic Neuroengineering and Symbolic Recirculation
      Modular Model for Parsing
  2.4 Summary and Discussion

3 A Hybrid Connectionist Scanning Understanding of Phrases
  3.1 Foundations of a Hybrid Connectionist Architecture
    3.1.1 Motivation for a Hybrid Connectionist Architecture
    3.1.2 The Computational Theory Level for a Scanning Understanding
    3.1.3 Constraint Integration
    3.1.4 Plausibility view
    3.1.5 Learning
    3.1.6 Subtasks of Scanning Understanding at the Computational Theory Level
    3.1.7 The Representation Level for a Scanning Understanding
  3.2 Corpora and Lexicon for a Scanning Understanding
    3.2.1 The Underlying Corpora
    3.2.2 Complex Phrases
    3.2.3 Context and Ambiguities of Phrases
    3.2.4 Organization of the Lexicon
  3.3 Plausibility Networks
    3.3.1 Learning Semantic Relationships and Semantic Context
    3.3.2 The Foundation of Plausibility Networks
    3.3.3 Plausibility Networks for Noun-Connecting Semantic Relationships
    3.3.4 Learning in Plausibility Networks
    3.3.5 Recurrent Plausibility Networks for Contextual Relationships
    3.3.6 Learning in Recurrent Plausibility Networks
  3.4 Summary and Discussion

4 Structural Phrase Analysis in a Hybrid Separated Model
  4.1 Introduction and Overview
  4.2 Constraints for Coordination
  4.3 Symbolic Representation of Syntactic Constraints
    4.3.1 A Grammar for Complex Noun Phrases
    4.3.2 The Active Chart Parser and the Syntactic Constraints
  4.4 Connectionist Representation of Semantic Constraints
    4.4.1 Head-noun Structure for Semantic Relationships
    4.4.2 Training and Testing Plausibility Networks with NCN-relationships
    4.4.3 Learned Internal Representation
  4.5 Combining Chart Parser and Plausibility Networks
  4.6 A Case Study
  4.7 Summary and Discussion

5 Structural Phrase Analysis in a Hybrid Integrated Model
  5.1 Introduction and Overview
  5.2 Constraints for Prepositional Phrase Attachment
  5.3 Representation of Constraints in Relaxation Networks
    5.3.1 Integrated Relaxation Network
    5.3.2 The Relaxation Algorithm
    5.3.3 Testing Relaxation Networks
  5.4 Representation of Semantic Constraints in Plausibility Networks
    5.4.1 Training and Testing Plausibility Networks with NPN-Relationships
    5.4.2 Learned Internal Representation
  5.5 Combining Relaxation Networks and Plausibility Networks
    5.5.1 The Interface between Relaxation Networks and Plausibility Networks
    5.5.2 The Dynamics of Processing in a Relaxation Network
  5.6 A Case Study
  5.7 Summary and Discussion

6 Contextual Phrase Analysis in a Hybrid Separated Model
  6.1 Introduction and Overview
  6.2 Towards a Scanning Understanding of Semantic Phrase Context
    6.2.1 Superficial Classification in Information Retrieval
    6.2.2 Skimming Classification with Symbolic Matching
  6.3 Constraints for Semantic Context Classification of Noun Phrases
  6.4 Syntactic Condensation of Phrases to Compound Nouns
    6.4.1 Motivation of Symbolic Condensation
    6.4.2 Condensation Using a Symbolic Chart Parser
  6.5 Plausibility Networks for Context Classification of Compound Nouns
    6.5.1 Training and Testing the Recurrent Plausibility Network
    6.5.2 Learned Internal Representation
  6.6 Summary and Discussion

7 Contextual Phrase Analysis in a Hybrid Integrated Model
  7.1 Introduction and Overview
  7.2 Constraints for Semantic Context Classification of Phrases
  7.3 Plausibility Networks for Context Classification of Phrases
    7.3.1 Training and Testing with Complete Phrases
    7.3.2 Training and Testing with Phrases without Insignificant Words
    7.3.3 Learned Internal Representation
  7.4 Semantic Context Classification and Text Filtering
  7.5 Summary and Discussion

8 General Summary and Discussion
  8.1 The General Framework of SCAN
  8.2 Analysis and Evaluation
    8.2.1 Evaluating the Problems
    8.2.2 Evaluating the Methods
    8.2.3 Evaluating the Representations
    8.2.4 Evaluating the Experiment Design
    8.2.5 Evaluating the Experiment Results
  8.3 Extensions of a Scanning Understanding
    8.3.1 Extending Modular Subtasks
    8.3.2 Extending Interactions
  8.4 Contributions and Conclusions

9 Appendix
  9.1 Hierarchical Cluster Analysis
  9.2 Implementation
  9.3 Examples of Phrases for Structural Phrase Analysis
  9.4 Examples of Phrases for Contextual Phrase Analysis

References
Index







Orders information
-----------------

            ISBN: 0 412 59100 6
            Pages: 190
            Figures: 56

            Price: 29.95 pounds sterling, 52.00 US dollars
            Credit cards: all major credit cards accepted by Chapman & Hall


            Please order from:

        --  Pam Hounsome
            Chapman & Hall
            Cheriton House
            North Way
            Andover
            Hants, SP10 5BE, UK
            England
 
            UK orders:
               Tel: 01264 342923 Fax: 01264 364418
            Overseas orders:
               Tel: +44 1264 342830 Fax: +44 1264 342761
 

            (Sister company in US)
        --  Peter Clifford
            Chapman & Hall Inc 
            One Penn Plaza
            41st Floor
            New York NY 10119
            USA
 
            Tel: 212 564 1060
            Fax: 212 564 1505

 
        --  or e-mail on: order at Chaphall.com




















From mario at physics.uottawa.ca  Tue Dec 13 09:01:44 1994
From: mario at physics.uottawa.ca (Mario Marchand)
Date: Tue, 13 Dec 94 10:01:44 AST
Subject: Neural net and PAC learning paper: errata
Message-ID: <9412131401.AA27027@physics.uottawa.ca>

About a week ago, I have sent you this message:


>The following paper, which was presented at the NIPS'94 conference,
>is available by anonymous ftp at:
>
>ftp://dirac.physics.uottawa.ca/usr2/ftp/pub/tr/marchand
>
>
>FileName: nips94.ps
>
>Title: Learning Stochastic Perceptrons Under k-Blocking Distributions
>
>Authors: Marchand M. and Hadjifaradji S.
>
>
>ALSO: you will find other papers co-authored by Mario Marchand in
>      this directory. The text file: Abstracts-mm.txt contains a
>      list of abstracts of all the papers.


In fact the right path for ftp is:

ftp://dirac.physics.uottawa.ca/pub/tr/marchand

and the file name is: nips94.ps.Z    (compressed)

sorry for this inconvenience and many thanks for those who have
kindly reported the mistakes.

- mario





   ----------------------------------------------------------------
  |  UUU            UUU            Mario Marchand                  |
  |  UUU            UUU            -----------------------------   |
  |  UUU       OOOOOOOOOOOOOOOO    Department of Physics           |
  |  UUU      OOO   UUU      OOO   University of Ottawa            |
  |   UUUUUUUUUUUUUUUU       OOO   150 Louis Pasteur street        |
  |           OOO            OOO   PO BOX 450 STN A                |
  |            OOOOOOOOOOOOOOOO    Ottawa (Ont) Canada K1N 6N5     |
  |                                                                |
  |   ***** Internet E-Mail: mario at physics.uottawa.ca **********   |
  |   ***** Tel: (613)564-9293 ------------- Fax: 564-6712 *****   |
   ----------------------------------------------------------------





From bisant at gl.umbc.edu  Tue Dec 13 20:21:48 1994
From: bisant at gl.umbc.edu (Mr. David Bisant)
Date: Tue, 13 Dec 1994 20:21:48 -0500
Subject: FLAME: WCNN'95 - science in the mud pits
In-Reply-To: <199412081651.LAA01963@garnet.cs.brandeis.edu>
Message-ID: <Pine.SGI.3.90.941213195922.11172B-100000@umbc8.umbc.edu>

On Thu, 8 Dec 1994, Jordan Pollack wrote:

> > 
> >Authors must submit 
> >registration payment with papers to be eligible for the early 
> >registration fee...Registration fee includes 1995 membership and a 
> >one (1) year subscription to  the Journal Neural Networks.
> >A $35 per paper processing fee must be enclosed 
> >in order for the paper to be refereed.
> 
> Reading fees are charged by agents for aspiring science fiction
> writers. Linking registration to submission is nothing more than an
> admission that all papers are to be accepted.  And a fee of $480 is
> paying a lot more than membership, subscription, and a few days of
> coffee breaks!
> 

The new fee structure for WCNN did not strike me as being unusually
high.  I also think it will help solve a problem I observed last 
year with a few authors not showing up for their talks, which wasted
valuable time slots for oral presentation and upset schedules.  

With interest in the neural network field diminishing, new problems
in conferencing appear.  I think their fee structure is a reasonable 
approach to solving these problems.

David Bisant
(no affiliation with the conference organization)


From cga at cc.gatech.edu  Tue Dec 13 21:35:22 1994
From: cga at cc.gatech.edu (Christopher G. Atkeson)
Date: Tue, 13 Dec 1994 21:35:22 -0500
Subject: About sequential learning (or interference)
Message-ID: <199412140235.VAA17037@lennon.cc.gatech.edu>

Memory-Based Learning is one approach to avoiding interference,
and is in part descended from nearest neighbor neural networks, or at
least neural networks with local representations.

Some of our references that will point you to other work as well:

Atkeson, C.G. (1992) "Memory-Based Approaches To Approximating
Continuous Functions." In: Casdagli, M., & Eubank, S. (eds.),
"Nonlinear Modeling and Forecasting" Addison Wesley, Redwood City, CA.

Atkeson, C.G. ``Using Local Models to Control Movement'',
Proceedings, Neural Information Processing Systems,
Dec 1989, Denver, Colorado.
In: Neural Information Processing Systems 2.
Morgan Kaufmann, 1990.

Schaal, S., & Atkeson, C.G. (1994).
"Robot Juggling: An Implementation of Memory-Based Learning."
IEEE Control Systems Magazine, vol. 15, no. 1, pp. 57-71.

Chris Atkeson
Stefan Schaal


From jlm at crab.psy.cmu.edu  Tue Dec 13 23:16:34 1994
From: jlm at crab.psy.cmu.edu (James L. McClelland)
Date: Tue, 13 Dec 94 23:16:34 EST
Subject: About sequential learning (or interference)
In-Reply-To: <Added.8ivEqJi00UdbIelU4r@andrew.cmu.edu> (message from DeLiang Wang on Mon, 12 Dec 1994 10:50:23 -0500)
Message-ID: <9412140416.AA17263@crab.psy.cmu.edu.psy.cmu.edu>


DeLiang Wang <dwang at cis.ohio-state.edu> writes:

> It was reported some time ago that multilayer perceptrons suffer the problem
> of so-called "catastrophic interference", meaning that later training will
> destroy previously acquired "knowledge" (see McCloskey & Cohen, Psychol. of
> Learning and Motivat. 24, 1990; Ratcliff, Psychol. Rev. 97, 1990). This seems
> to be a serious problem, if we want to use neural networks both as a stable
> knowledge store and a long-term problem solver.

The brain appears to have solved this problem by storing new
information in a special associative memory system in the hippocampus.
According to this view, cortical (and some other non-hippocampal)
systems learn slowly, using what I call 'interleaved learning'.
Weights are adjusted a small amount after each experience, so that the
overall direction of weight change is governed by the structure
present in the ensemble of events and experiences.  New material can
be added to such a memory without catastrophic intereference if it is
added slowly, interleaved with ongoing exposure to other events and
experiences.  This is too slow for the demands placed on memory by the
world.  To allow rapid learning of new material without catastrophic
interference, new material is initially stored in the hippocampus,
where sparse, conjunctive representations are used to minimize
interference with other memories.  Reinstatement of these patterns
occurs in task-relevant situations via bi-directional connections
between hippocampus and cortex together with pattern completion within
the hippocampal formation.  Interestingly, it appears that
reinstatement occurs in off-line situations as well -- most notably,
during sleep.  Each reinstatement provides an opportunity for the
neocortex to learn; but these reinstatements occur interleaved with
other ongoing events and experiences, and weight changes are small on
each reinstatement, so that the neocortex learns the new material via
interleaved learning.  This theory is consistent with a lot of data at
this point, including of course the basic fact that bilateral removal
of the hippocampus leads to a profound deficit in the ability to form
arbitrary new memories rapidly.  Among the most important additional
findings are 1) the phenomenon of temporally graded retrograde amnesia
--- the finding that damage to the hippocampus can produce a selective
deficit for recently acquired memories leaving memories acquired
several weeks or months earlier intact --- and 2) the finding that
neurons in the hippocampus that are coactive during a particular
behavior appear to be coactive during sleep following the behavior, as
if the patterns of activity that were present during behavior are
reactivated during sleep.

I announced a technical report discussing the above ideas last March.
In that report, we focused on why it makes sense from a connectionist
point of view that the system should be organized as described above.
A reprint of the announcement (sans abstract) appears below.  This TR
is currently under revision for publication; the revised version will
contain a fuller presentation of the physiological evidence than the
present version, and I will announce it on connectionists when it
becomes available.

========================================================================
Technical report announcement:
------------------------------------------------------------------------

             Why there are Complementary Learning Systems
                  in the Hippocampus and Neocortex:
             Insights from the Successes and Failures of
             Connectionist Models of Learning and Memory

    James L. McClelland, Bruce L. McNaughton & Randall C. O'Reilly
        Carnegie Mellon University & The University of Arizona

                    Technical Report PDP.CNS.94.1
                             March, 1994

=======================================================================

Retrieval information:

unix> ftp 128.2.248.152                 # hydra.psy.cmu.edu
Name: anonymous
Password: <email address>
ftp> cd pub/pdp.cns
ftp> binary
ftp> get pdp.cns.94.1.ps.Z
ftp> quit
unix> zcat pdp.cns.94.1.ps.Z | lpr      # or however you print postscript

NOTE:  

The compressed file is 306994 bytes long.
Uncompressed, the file is 840184 byes long.

The printed version is 63 total pages long.

For those who do not have FTP access, physical copies can be requested from
Barbara Dorney <bd1q+ at andrew.cmu.edu>.  Ask for the report by title or
pdp.cns technical report number.






From bogner at eleceng.adelaide.edu.au  Tue Dec 13 23:22:37 1994
From: bogner at eleceng.adelaide.edu.au (Robert E. Bogner)
Date: Wed, 14 Dec 1994 14:52:37 +1030
Subject: [dwang@cis.ohio-state.edu: About sequential learning (or interference)]
Message-ID: <9412140422.AA16796@augean.eleceng.adelaide.edu.au>


From dwang at cis.ohio-state.edu  Mon Dec 12 10:50:23 1994
From: dwang at cis.ohio-state.edu (DeLiang Wang)
Date: Mon, 12 Dec 1994 10:50:23 -0500
Subject: About sequential learning (or interference)
Message-ID: <mailman.717.1149591322.29955.connectionists@cs.cmu.edu>

1. Visakan Kadirkamanathan wrote a dissertation on "Sequential
Learning in Artifician neural networks for PhD at U. of Cambridge,
October 1991. there were some conference papers undeer the same author.

2. C. J. Wallace, "Autonomous Rehearsal: Stability and plasticity in a
multilayer perceptron", project report, Dept. of EEE, Univ. of
Adelaide, October 1992  " . . a psychologically-motivated idea for
mainaining the original learning of a classifier neural network by
getting it to reinforce associations formed from confidently
classified random inputs. . ."

Robert E. Bogner

--------------------------------------------------------------------------
e_mail         bogner at eleceng.adelaide.edu.au             WATCH THIS
                                                            SPACE
Reliable mail: Robert E. Bogner
               Professor of Electrical Engineering,       
	       The University of Adelaide,          	  
	       Adelaide 5005,                       	  
	       SOUTH AUSTRALIA                      	  

Phone 	+61 8 303 5589 (answering machine all hours)    
Fax  	+61 8 303 4360  OR +61 8 224 0464
Home phone +61 8 332 5549
Time GMT+9h30m in northern hemisphere summer
     GMT+10h30m in northern hemisphere winter
==========================================================================


From sef+ at cs.cmu.edu  Wed Dec 14 00:03:50 1994
From: sef+ at cs.cmu.edu (Scott E. Fahlman)
Date: Wed, 14 Dec 94 00:03:50 EST
Subject: About sequential learning (or interference) 
In-Reply-To: Your message of Tue, 13 Dec 94 21:35:22 -0500.
             <199412140235.VAA17037@lennon.cc.gatech.edu> 
Message-ID: <mailman.718.1149591322.29955.connectionists@cs.cmu.edu>


Yet another approach to incremental learning can be seen in the Cascade
Correlation algorithm.  This creates hidden units (or feature detectors, if
you prefer) one by one, and freezes each one after it is created.  The
weights between these features and the outputs remain plastic and continue
to be trained.

This means that if you train a net on training set A and then switch to a
different training set B, you may build some new hidden units for task B,
but the hidden units created for task A are not cannibalized.  The A units
may be used in task B, either directly or as inputs to new hidden units,
but they remain unchanged and available.  As task B is trained, the output
weights change, so performance on task A will generally decline, but it can
come back very quickly if task A is re-trained or if you now train with a
combined A+B task.  The point is that the time-consuming part of learning
is in creating the hidden units, and these are retained once they are
learned.

My Recurrent Cascade-Correlation paper has an example of this.  This
recurrent net can be trained to recognize a temporal sequence of 1's and
0's as Morse code.  If you train all 26 letter-codes as a single training
set, the network will learn the task, but learning is faster and the
resulting net is smaller if you break the training set up into "lessons" of
increasing difficulty: first train on the shortest codes, then the
medium-length ones, then the longest ones, and finally on the whole set
together.  This is reminiscent of the modular networks explored by Waibel
and his colleagues for large speech tasks, but in this case you just have
to chop the training set up into modules -- the network architecture takes
care of itself.

References (these are also available online on Neuroprose, among other
places):

Scott E. Fahlman and Christian Lebiere (1990) "The Cascade-Correlation
Learning Architecture", in {\it Advances in Neural Information Processing
Systems 2}, D. S. Touretzky (ed.), Morgan Kaufmann Publishers, Los Altos
CA, pp. 524-532.

Scott E. Fahlman (1991) "The Recurrent Cascade-Correlation Architecture" in
{\it Advances in Neural Information Processing Systems 3}, R. P. Lippmann,
J. E. Moody, and D. S. Touretzky (eds.), Morgan Kaufmann Pulishers, Los
Altos CA, pp. 190-196.

-- Scott

===========================================================================
Scott E. Fahlman			Internet:  sef+ at cs.cmu.edu
Principal Research Scientist		Phone:     412 268-2575
School of Computer Science              Fax:       412 268-5576 (new!)
Carnegie Mellon University		Latitude:  40:26:46 N
5000 Forbes Avenue			Longitude: 79:56:55 W
Pittsburgh, PA 15213			Mood:      :-)
===========================================================================


From pollack at cs.brandeis.edu  Wed Dec 14 14:45:33 1994
From: pollack at cs.brandeis.edu (Jordan Pollack)
Date: Wed, 14 Dec 1994 14:45:33 -0500
Subject: Apology for Flame, and Survey 
In-Reply-To: <Pine.SGI.3.90.941213195922.11172B-100000@umbc8.umbc.edu> (bisant@gl.umbc.edu)
Message-ID: <199412141945.OAA10961@garnet.cs.brandeis.edu>

I flamed about some of the policies of WCNN, and David Bisant wrote:

> new problems
>in conferencing appear.  I think their fee structure is a reasonable 
>approach to solving these problems.

I obviously over-reacted, since I agree that there is also a more
positive reading of these policies which I didnt consider at the time -
that they are innovations that that the organization is using to cope
with several problems, such as:

a) filtering out computer-generated articles,
b) authors not showing up to deliver talks or posters,
c) expense of supporting travel costs for both plenary 
   speakers and conference organizers, and
d) declining membership/subscription revenues.

It was the simultaneous appearance of all these novel structures in
one place which raised a red flag, and led me to breathe fire.  I am
not usually so rude, and now I owe MANY colleagues apologies for my
outburst.

First, I'm very sorry for linking the two completely separate
societies - IEEE and INNS, and for denigrating the papers of
individuals who have published their work in these refereed meetings
over the years. Also, I want to apologize to the dozens of major
scientists involved in organizing and governing INNS, for sullying
their good names, and finally to Mike Cohen and my other new neighbors
at Boston Univ, who are merely using their good offices to promote the
meeting.

I hope the society will eventually forgive my unsociable outburst.

However, I do believe the $35 reviewing fee is a serious and big
issue, as I have never seen a precedent in any branch of science. And
while there are real costs to reviewing which are bundled into
the costs of meetings, this "unbundling" is a very slippery slope
to walk. If such fees catch on in conferences, they might catch on
universally, from journals to funding agencies to faculty search
committees, and this would dramatically change the "social contract,"
I fear, with many more negative than positive results!  But perhaps
even in this sub-issue I am over-reacting. Since the list is too large for
a discussion, I would like to perform an informal survey, which I
will then condense and post the results:

1) What do you think about review fees for scientific conferences?

2) What about review fees for scientific journals?

3) What about review fees for grant applications to NSF/NIH and other agencies?

4) What about review fees for faculty and postdoc job applications?

5) What do you predict as the ramifications of changing the current
scientific free review system to a fee-based review system?

Jordan Pollack                           Associate Professor
Computer Science Department              Center for Complex Systems
Brandeis University                      Phone: (617) 736-2713/fax 2741
Waltham, MA 02254                        email: pollack at cs.brandeis.edu




From white at pwa.acusd.edu  Wed Dec 14 14:44:50 1994
From: white at pwa.acusd.edu (Ray White)
Date: Wed, 14 Dec 1994 11:44:50 -0800 (PST)
Subject: FLAME: WCNN'95 - science in the mud pits
In-Reply-To: <199412081651.LAA01963@garnet.cs.brandeis.edu>
Message-ID: <Pine.3.89.9412141127.A16497-0100000@pwa.acusd.edu>

I'm not thrilled with the "reading fee" either, but $35 plus early
member registration @ $170 is pretty low for a conference these days;
certainly far less than the maximum registration fee of $480 that
Jordan Pollack mentioned.

As far as "sorting papers by fame" goes, for those of us who are not
famous, it is a good thing that there are conferences where papers
are merely "sorted by fame", when there are also elitist conferences
where fame is more likely a requirement for acceptance. (Did anyone
read NIPS into that sentence?) A conference is remembered for the quality
of presented papers, not for the quality of rejected papers.

Ray H. White			|  Depts. of Physics and Computer Science
<white at teetot.acusd.edu>	|  University of San Diego
619/260-4244 or 619/260-4627	|  5998 Alcala Park,  San Diego, CA 92110







From jose at scr.siemens.com  Wed Dec 14 15:20:45 1994
From: jose at scr.siemens.com (Stephen Hanson)
Date: Wed, 14 Dec 1994 15:20:45 -0500 (EST)
Subject: About sequential learning (or interference)
In-Reply-To: <199412140235.VAA17037@lennon.cc.gatech.edu>
References: <199412140235.VAA17037@lennon.cc.gatech.edu>
Message-ID: <YivpCRu1GEMnA0vvJt@tractatus.siemens.com>


of course avoiding "interference" is
another way of preventing generalization.

As usual there is a tradeoff here.

Steve




Stephen J. Hanson, Ph.D.
Head, Learning Systems Department
SIEMENS Research 
755 College Rd. East
Princeton, NJ 08540



From lautrup at connect.nbi.dk  Thu Dec 15 11:56:25 1994
From: lautrup at connect.nbi.dk (Benny Lautrup)
Date: Thu, 15 Dec 94 11:56:25 MET
Subject: New preprint: "Massive Weight Sharing: A Cure for ..."
Message-ID: <mailman.719.1149591322.29955.connectionists@cs.cmu.edu>

Subject: Paper available: "Massive Weight Sharing: A Cure for ..."
Date: August 29, 1994

FTP-host: connect.nbi.dk
FTP-file: neuroprose/lautrup.massive.ps.Z

----------------------------------------------

The following paper is now available:

Massive Weight Sharing: A Cure for Extremely Ill-posed Learning [12 pages]

B. Lautrup, L.K. Hansen, I. Law, N. Moerch, C. Svarer and S.C. Strother
CONNECT, The Niels Bohr Institute, University of Copenhagen, Denmark

Presented at the workshop on Supercomputing in Brain Research: From Tomography
to Neural Networks, HLRZ, Juelich, Germany, November 21-23, 1994

Abstract:
In most learning problems, adaptation to given examples is  well-posed  because
the number of examples far exceeds the number of  internal  parameters  in  the
learning machine. Extremely ill-posed learning problems are, however, common in
image and spectral analysis. They are characterized by a vast number of  highly
correlated inputs, e.g. pixel or pin values, and a modest number  of  patterns,
e.g. images or spectra. In this paper we show, for the case of  a  set  of  PET
images differing only in the values of  one  stimulus  parameter,  that  it  is
possible to train a neural network to learn the underlying rule  without  using
an excessive number of network weights or large amounts of computer  time.  The
method is based upon the observation that the standard learning rules  conserve
the subspace spanned by the input images.

Please do not reply directly to this message.

-----------------------------------------------

FTP-instructions:

unix> ftp connect.nbi.dk (or 130.225.212.30)
ftp> Name: anonymous
ftp> Password: your e-mail address
ftp> cd neuroprose
ftp> binary
ftp> get lautrup.massive.ps.Z
ftp> quit
unix> uncompress lautrup.massive.ps.Z
                                             
-----------------------------------------------

Benny Lautrup,
Computational Neural Network Center (CONNECT)
Niels Bohr Institute
Blegdamsvej 17
2100 Copenhagen
Denmark

Telephone: +45-3532-5200
Direct:    +45-3532-5358
Fax:       +45-3142-1016
e-mail:    lautrup at connect.nbi.dk


From pah at unixg.ubc.ca  Thu Dec 15 14:03:03 1994
From: pah at unixg.ubc.ca (Phil A. Hetherington)
Date: Thu, 15 Dec 1994 11:03:03 -0800 (PST)
Subject: sequential learning
In-Reply-To: <9412140416.AA17263@crab.psy.cmu.edu.psy.cmu.edu>
Message-ID: <Pine.SUN.3.91.941215105220.19696A-100000@unixg.ubc.ca>

A couple notes on interfernce. First, most of Jay's idea is still simply
theory, a good theory, but there is still little direct evidence that the
hippocampus does either pattern separation or pattern completion.  Most of
the pattern completion idea comes from Marr's conjectures on the
architecture of CA3 in hippocampus.  Second (to Hanson), interference is
not necessarily the flip side of generalization.  This is intuitive, but you
can reduce one without affecting the other.  See Lewandowsky's stuff on
this .  Next, whether the interference observed in connectionist nets is a
problem or not depends on your needs.  Is the net to model human learning
or to solve a particular engineering problem.  If the former, then see
McRae and Hetherington (1993)--it is not clear that realisticly designed
nets suffer from catastrophic interference.  Finally, it appears the the
bigger problem is to reduce interference without eliminating the
distinction between old and new items.  This was Ratcliff's main concern,
and is addressed by Sharkey. 

See also:

French, R.M. (1991). Proceedings of the Thirteenth Annual Conference of
  the Cognitive Science Society, 173-178. 
French, R.M. (1992). Connection Science, 4, 3-4, 365-377. 
Hetherington, P.A. (1990).  Neural Network Review, 4(1), 27-29. 
Hetherington, P.A. (1991).  The Sequential Learning Problem in 
  Connectionist Networks.  Unpublished Masters Thesis, McGill
  University, Montreal, Quebec. 
Hetherington, P.A., & Seidenberg, M.S. (1989).  Proceedings of the 
  Eleventh Annual Conference of the Cognitive Science Society, 26-33. 
Hinton, G.E., & Plaut, D.C. (1987).  Proceedings of the Ninth Annual 
  Conference of the Cognitive Science Society, 177-186.
Kortge, C.A. (1990). Proceedings of the Twelfth Annual Conference of the
  Cognitive Science Society, 764-771. 
Krushke, J.K. (1992).  Psychological Review, 99, 22-44. 
Krushke, J.K. (1993).  Connection Science, 5, 3-36.
Lewandowsky, S. (1991).  In W.E. Hockley & S, Lewandowsky (Eds.), Relating
  theory and data: Essays on human memory in honor of Bennet B. Murdock.
McRae, K., & Hetherington, P.A. (1993). Proceedings of the Fifteenth
  Annual Conference of the Cognitive Science Society, 723-728. 
Murre, J.M.J. (1992).  Learning and categorization in modular neural 
  networks.  Lawrence Erlbaum. 
Sharkey, N.E., & Sharkey, A.J.C. Understanding Catastrophic
  Interference in neural Nets.  Technical Report, Department of Computer
  Science, University of Sheffield, U.K. 
Sloman, S.A., & Rumelhart, D.E. (1992).  In A. Healy, S.M. Kosslyn, & 
  R.M. Shiffrin (Eds.), From learning theory to cognitive processes: Essays 
  in honor of William K. Estes. 





From french at cogsci.indiana.edu  Thu Dec 15 14:28:01 1994
From: french at cogsci.indiana.edu (Bob French)
Date: Thu, 15 Dec 94 14:28:01 EST
Subject: Catastrophic forgetting and sequential learning
Message-ID: <mailman.720.1149591322.29955.connectionists@cs.cmu.edu>


Below I have indicated a number of references on current work on the
problem of sequential learning in connectionist networks.  All of these
papers address the problem of catastrophic interference, which may
result when a previously trained connectionist network attempts to
learn new patterns.
   The following commented list is by no means complete.  In
particular, no mention is made of convolution-correlation models with
their associated connectionist networks.  Nonetheless, I hope it
might prove to be a useful introduction to people interested in
knowing a bit more about the subject.

				Bob French
                                french at cogsci.indiana.edu


----------------------------------------------------------------------------

                   Recent Work in Catastrophic Interference 
                          in Connectionist Networks


The two papers that really kicked off research in this area were:

McCloskey, M. & Cohen, N. (1989)  "Catastrophic interference in connections
	networks: the sequential learning problem" The Psychology of 
	Learning and Motivation, 24, 109-165.

Ratcliff, R (1990) "Connectionist models of recognition memory: constraints
	imposed by learning and forgetting functions" Psychological Review,
	97, 285-308


Hetherington and Seidenberg very early suggested an Ebbinghaus-like
"savings" measure of catastrophic interference and, based on this,
they concluded that catastrophic interference wasn't really as much of
a problem as had been thought. While the problem has, in fact, been
subsequently confirmed to be quite serious, the "savings" measure they
proposed is still widely used to measure the extent of forgetting.

Hetherington, P.. & Seidenberg, M. (1989) "Is there 'catastrophic
	interference' in connectionist networks?"  Proceedings of the
	11th Annual Conference of the Cognitive Science Society. 
	Hillsdale, NJ: Erlbaum, 26-33.


Kortge was one of the first to propose a solution to this problem,
using what he called "novelty vectors".

Kortge, C. (1990) "Episodic Memory in Connectionist Networks" Proceedings 
	of the 12th Annual Conference of the Cognitive Science Society. 
	Hillsdale, NJ: Erlbaum, 764-771


Sloman and Rumelhart also developed a technique called "episodic gating"
designed to reduce the severity of the problem.

Sloman, S. & Rumelhart, D., (1991) "Reducing interference in distributed
	memories through episodic gating" In Healy, Kosslyn, & Shiffrin (eds.)
	Essays in Honor of W. K. Estes.


In 1991 French suggested that catastrophic interference might be the 
inevitable price you pay for the advantages of fully distributed
representations (in particular, generalization).  He suggested a way
of dynamically producing "semi-distributed" hidden-layer
representations to reduce the effect of catastrophic interference.

French, R. (1991) "Using semi-distributed representations to overcome 
	catastrophic forgetting in connectionist networks" in Proceedings 
	of the 13th Annual Conference of the Cognitive Science Society. 
	Hillsdale, NJ: Erlbaum, 173-178.

A more detailed article presenting the same techique, called
activation sharpening, appeared in Connection Science.

French, R. (1992) "Semi-distributed Representations and Catastrophic
	Forgetting in Connectionist Networks", Connection Science,
	Vol. 4: 365-377.

In a more recent paper (1994), French presented a technique called
context biasing, which again dynamically "massages" hidden layer
representations based on the "context" of other recently learned
exemplars.  The goal of this technique is to produce hidden-layer
representations that are simultaneously well distributed and
orthogonal.

French, R. (1994) "Dynamically constraining connectionist networks to
	produce distributed, orthogonal representations to reduce
	catastrophic interference" in Proceedings of the 16th Annual 
	Conference of the Cognitive Science Society. Hillsdale, NJ: 
	Erlbaum, 335-340

Finally, still in this vein, French proposed (1993) at the NIPS-93
workshop on catastrophic interference a dynamic system of two
interacting networks working in tandem, one storing prototypes, the
other doing short-term learning of new exemplars. For a "real brain"
justification for this type of architecture see McClelland,
McNaughton, and O'Reilly (1994), below.  (A full paper on this tandem
network architecture will be available at the beginning of next year.)
This technique is discussed very briefly in:

French, R. (1994) "Catastrophic interference in connectionist networks:
	Can it be predicted, can it be prevented?" Neural Information
	Processing Systems - 6, Cowan, Tesauro, Alspector (eds.) 
	San Francisco, CA: Morgan Kaufmann. 1176-1177.


Steve Lewandowsky has also been very active in this area.  He
developed a simple technique in 1991 that focused on producing
orthogonalization at the input layer rather than the hidden layer.
This "symmetric vectors" technique is discussed in:

Lewandowsky, S. & Shu-Chen Li (1993) "Catastrophic Interference in
	Neural Networks: Causes, solutions and data" in New 
	Perspectives on interference and inhibition in cognition.
	Dempster & Brainerd (eds.) New York: NY: Academic Press.

Lewandowsky, S (1991) "Gradual unlearning and catastrophic
	interference: a comparison of distributed architectures.
 	In Hockley & Lewandowsky (eds.) Relating theory and
	data: Essays on human memory in honor of Bennet B.
	Murdock. (pp. 445-476).  Hillsdale, NJ: Lawrence Erlbaum.

and in an earlier University of Oklahoma psychology department
technical report:

Lewandowsky, S. (1993) "On the relation between catastrophic
	interference and generalization in connectionist networks"


In 1993 McRae and Hetherington published a study using pre-training to
eliminate catastrophic interference.

McRae, K. & Hetherington, P. (1993) "Catastrophic interference is 
	eliminated in pretrained networks" in Proceedings of the
	15th Annual Cognitive Science Society.  Hillsdale, NJ: 
	Erlbaum. 723-728


John Kruschke discussed the problem of catastrophic forgetting at
length in the context of his connectionist model, ALCOVE, and showed
the extent to which and under what circumstances this model is not
subject to catastrophic forgetting.

Kruschke, J. (1993) "Human category learning: implications for
	backpropagation models", Connection Science, Vol. 5, No. 1, 


Jacob Murre has also examined how his model, CALM, performs on
the sequential learning problem.    See, in particular:

Murre, J. Learning and Categorization in Modular neural networks
	Hillsdale, NJ: Lawrence Erlbaum. 1992. (see esp. ch. 7.4)


It is to be noted that both ALCOVE and CALM rely, at least in part,
on reducing the distributedness of their internal representations in
order to achieve improved performance on the problem of catastrophic
interference.


A 1994 article (in press) by Anthony Robins presents a novel
technique, called "pseudorehearsal", whereby "pseudoexemplars" that
reflect prior learning are added to the new data set to learned in
order to reduce catastrophic forgetting.

Robins, A. "Catastrophic forgetting, rehearsal, and pseudorehearsal",
	University of Otago (New Zealand) computer science technical
	report.  (copies:  coscavr at otago.ac.nz)


Tetewsky, Shultz & Buckingham (1994) demonstrate the improvements that
result from using Fahlman's cascade-correlation learning algorithm.

Tetewsky, S., Shultz, T. and Buckingham, D. "Assessing interference and
	savings in connectionist models of human recognition memory"
	Department of Psychology TR, McGill University, Montreal.
	(presented at 1994 Meeting of the Psychonomic Society).


Sharkey & Sharkey (1993) discussed the relation between the problem of
interference and discrimination in connectionist networks.  They
conclude that sequentially trained networks using backprop will
unavoidably suffer from one or the other problem.  I am not aware if
there is a final version of this paper in print yet, but Noel Sharkey
is currently at University of Sheffield, Dept. of Computer Science,
Sheffield.  n.sharkey at dcs.shef.ac.uk

Sharkey, N. & Sharkey, A., "An interference-discrimination tradeoff
	in connectionist models of human memory"  


McClelland, McNaughton, & O'Reilly issued a CMU technical report
earlier this year (1994) in which they discuss the phenomenon of
catastrophic interference in the context of the "real world", i.e.,
the brain.  They suggest that the complementary learning systems in
the hippocampus and the neocortex might be the brain's way of 
overcoming the problem.  They argue that this dual system provides 
a means not only of rapidly acquiring new information, but also of 
storing well-learned information as prototypes.

McClelland, J., McNaughton, B., & O'Reilly, R. "Why there are
	complementary learning systems in the hippocampus and 
	neocortex:  Insights from the successes and failures
	of connectionist models of learning and memory"  CMU
	Tech report: PNP.CNS.94.1, March 1994.

-------------------------------------------------------------------------




From marney at ai.mit.edu  Thu Dec 15 18:19:23 1994
From: marney at ai.mit.edu (Marney Smyth)
Date: Thu, 15 Dec 94 18:19:23 EST
Subject: NNSP95 : Call for Papers
Message-ID: <9412152319.AA00894@motor-cortex>

**********************************************************************
*                  1995 IEEE WORKSHOP ON			     *
*   								     *
*           NEURAL NETWORKS FOR SIGNAL PROCESSING		     *
*								     *
*     August 31 -- September 2, 1995, Cambridge, Massachusetts, USA  *
*          Sponsored by the IEEE Signal Processing Society           *
*       (In cooperation with the IEEE Neural Networks Council)       *
*								     *
**********************************************************************


                  FIRST ANNOUNCEMENT AND CALL FOR PAPERS
 
Thanks to the sponsorship of IEEE Signal Processing Society, the
co-sponsorship of IEEE Neural Network Council, the fifth of a series
of IEEE Workshops on Neural Networks for Signal Processing will be
held at the Royal Sonesta Hotel, Cambridge, Massachusetts, on Thursday
8/31 -- Saturday 9/2, 1995. Papers are solicited for, but not limited
to, the following topics:

++ APPLICATIONS:
Image, speech, communications, sensors, medical, adaptive
filtering, OCR, and other general signal processing and pattern
recognition topics.

++ THEORIES:
Generalization and regularization, system identification, parameter
estimation, new network architectures, new learning algorithms, and
wavelets in NNs. 

++ IMPLEMENTATIONS:
Software, digital, analog, hybrid technologies, parallel processing.


Prospective authors are invited to submit 5 copies of extended summaries
of no more than 6 pages. The top of the first page of the summary should
include a title, authors' names, affiliations, address, telephone and
fax numbers and e-mail address if any. Camera-ready full papers
of accepted proposals will be published in a hard-bound volume by IEEE
and distributed at the workshop.  For further information, please
contact Marney Smyth (Tel.) (617)-253-0547, (Fax) (617)-253-2964
(e-mail) marney at ai.mit.edu. 

We plan to use the World Wide Web (WWW) for posting further announcements 
on NNSP95 such as: submitted papers status, final program, hotel information 
etc. You can use MOSAIC and access URL site: http://www.cdsp.neu.edu.
If you do not have access to WWW use anonymous ftp to site ftp.cdsp.neu.edu
and look under the directory /pub/NNSP95.

Please send paper submissions to:

Prof. Elias S. Manolakos 
IEEE  NNSP'95
409 Dana Research Building 
Electrical and Computer Engineering Department 
Northeastern University, Boston, MA 02115, USA 

Phone: (617) 373-3021,  Fax: (617) 373-4189                  

*******************
* IMPORTANT DATES *
*******************

Extended summary received by:            February 17

Notification of acceptance:              April 21

Photo-ready accepted papers received by: May 22 

Advanced registration received before:   June 2 

                                    
GENERAL CHAIRS

Federico Girosi
Center for Biological and Computational Learning, MIT
girosi at ai.mit.edu

John Makhoul
Bolt, Beranek and Newman
makhoul at bbn.com


PROGRAM CHAIR

Elias S. Manolakos
Communications and Digital Signal Processing (CDSP)
Northeastern University
elias at cdsp.neu.edu

FINANCE CHAIR			LOCAL ARRANGEMENTS

Judy Franklin                   Mary Pat Fitzgerald, MIT
GTE Laboratories		marypat at ai.mit.edu
jfranklin at gte.com

PUBLICITY CHAIR			PROCEEDINGS CHAIR

Marney Smyth, MIT		Elizabeth J. Wilson
marney at ai.mit.edu		Raytheon Co. 			
				email: bwilson at sud2.ed.ray.com


TECHNICAL PROGRAM COMMITTEE

Joshua Alspector		John Makhoul
Alice Chiang			Elias Manolakos
A. Constantinides		P. Mathiopoulos 
Lee Giles			Nahesan Niranjan
Federico Girosi			Tomaso Poggio
Lars Kai Hansen 		Jose Principe
Yu-Hen Hu			Wojtek Przytula
Jenq-Neng Hwang			John Sorensen
Bing-Huang Juang		Andreas Stafylopatis
Shigeru Katagiri		John Vlontzos
George Kechriotis		Raymond Watrous
Stephanos Kollias		Christian Wellekens
Sun-Yuan Kung			Ron Williams
Gary M. Kuhn			Barbara Yoon
Richard Lippmann		Xinhua Zhuang 





From Dave_Touretzky at DST.BOLTZ.CS.CMU.EDU  Fri Dec 16 03:22:31 1994
From: Dave_Touretzky at DST.BOLTZ.CS.CMU.EDU (Dave_Touretzky@DST.BOLTZ.CS.CMU.EDU)
Date: Fri, 16 Dec 94 03:22:31 EST
Subject: elitism at NIPS
Message-ID: <1394.787566151@DST.BOLTZ.CS.CMU.EDU>

The folks involved in running NIPS are aware that in the past, some people
have felt the conference was biased against outsiders.  As the program
chair for NIPS*94, I want to mention some steps we took to reduce that
perception:

* This year we required authors to submit full papers, not just abstracts.
  That made it harder for "famous" people to get by on their reputation
  alone, and easier for "non-famous" people to get good papers in.

* This year we required reviewers to provide real feedback to all authors.
  In order to make this less painful, we completely redesigned the review
  form (it's publicy accessible via the NIPS home page) and, for the first
  time, accepted reviews by email.  Everyone liked getting comments from
  the reviewers, and authors whose papers were not accepted understood why.

* We continue to recruit new people for positions in both the organizing
  and program committees.  It's not the same dozen people year after year.
  We also have a large reviewer pool:  176 people served as reviewers
  this year.

* We tend to bring in "outsiders" as our invited speakers, rather than
  the usual good old boys.  This year's invited speakers included Francis
  Crick of the Salk Insititute, Bill Newsome (a neuroscientist from 
  Stanford), and Malcolm Slaney (a signal processing expert formerly with
  Apple and now at Interval Research.)  None had been to NIPS before.

The fundamental limits on NIPS paper acceptances are that (1) we're
committed to a single-track conference, and (2) we only have room for 138
papers in the proceedings.  Therefore the reviewing process has to be
selective.

The acceptance rate this year was 33%; it was 25% in past years.  The
change is due to fewer submissions, not more acceptances.  Requiring full
papers was probably the cause of the drop in submissions.  The quality of
the accepted papers has remained high.

There are other ways to participate in NIPS besides writing a paper.  We
have a very successful workshop program, where lots of intense interaction
takes place between people with similar interests.  Many people give talks
at the workshops and not at the conference.  (I did that this year.)  NIPS
issues a call for workshop proposals at about the same time as the call for
papers.  Consider organizing a workshop at NIPS*95, or at least
participating in one.  You do not even have to register for the conference;
workshop registration is entirely separate.

The URL for the NIPS home page appears below.  Besides the review form,
you'll also find formatting instructions for paper submissions.  Authors of
accepted NIPS*94 papers will find instructions for how to submit their
final camera-ready copy, which is due January 23.

  http://www.cs.cmu.edu:8001/afs/cs/project/cnbc/nips/NIPS.html

-- Dave Touretzky
   NIPS*94 Program Chair


From P.McKevitt at dcs.shef.ac.uk  Fri Dec 16 05:55:02 1994
From: P.McKevitt at dcs.shef.ac.uk (Paul Mc Kevitt)
Date: Fri, 16 Dec 94 10:55:02 GMT
Subject: AISB-95 workshop call for papers
Message-ID: <mailman.721.1149591322.29955.connectionists@cs.cmu.edu>


<<FIRST CALL FOR PAPERS>>  <<FIRST CALL FOR PAPERS>>  <<FIRST CALL FOR PAPERS>>
<<FIRST CALL FOR PAPERS>>  <<FIRST CALL FOR PAPERS>>  <<FIRST CALL FOR PAPERS>>

                            Advance Announcement
  
                  FIRST CALL FOR PAPERS AND PARTICIPATION

                            AISB-95 Workshop on

                             REACHING FOR MIND:
                      FOUNDATIONS OF COGNITIVE SCIENCE

                            April 3rd/4th 1995

                                  at the
       The Tenth Biennial Conference on AI and Cognitive Science (AISB-95)
                   (Theme: Hybrid Problems, Hybrid Solutions)

                                 Halifax Hall
                           University of Sheffield
                              Sheffield, England
                    (Monday 3rd -- Friday 7th April 1995)
 
                          Society for the Study of
        Artificial Intelligence and Simulation of Behaviour (SSAISB)

                                   Chair:
                              Sean O Nuallain                              

                 Dublin City University, Dublin, Ireland &
                 National Research Council, Ottawa, Canada

                                 Co-Chair:
                               Paul Mc Kevitt

                       Department of Computer Science
                      University of Sheffield, England



WORKSHOP COMMITTEE:

John Barnden      (New Mexico State University, NM, USA)
Istvan Berkeley   (University of Alberta, Canada)
Mike Brady        (Oxford, England)
Harry Bunt        (ITK, Tilburg, The Netherlands)
Peter Carruthers  (University of Sheffield, England) 
Daniel Dennett    (Tufts University, USA)
Eric Dietrich     (SUNY Binghamton, NY, USA)
Jerry Feldman     (ICSI, UC Berkeley, USA)
John Frisby       (University of Sheffield, England)
Stevan Harnad     (University of Southampton, England)
James Martin      (University of Colorado at Boulder, CO, USA)
John Macnamara    (McGill University, Canada)
Mike McTear       (Universities of Ulster and Koblenz, Germany)
Ryuichi Oka       (RWC P, Tsukuba, Japan)
Jordan Pollack    (Ohio State University, OH, USA)
Zenon Pylyshyn    (Rutgers University, USA)
Ronan Reilly      (University College, Dublin, Ireland)
Roger Schank      (ILS, Illinois)
NNoel Sharkey      (University of Sheffield, England)
Walther v.Hahn    (University of Hamburg, Germany)
Yorick Wilks      (University of Sheffield, England)


WORKSHOP DESCRIPTION

The assumption underlying this workshop is that Cognitive Science (CS)
is in crisis.  The crisis manifests itself, as exemplified by the
recent Buffalo summer institute, in a complete lack of consensus among
even the biggest names in the field on whether CS has or indeed should
have a clearly identifiable focus of study; the issue of identifying
this focus is a separate and more difficult one. Though academic
programs in CS have in general settled into a pattern compatible with
classical computationalist CS (Pylyshyn 1984, Von Eckardt 1993),
including the relegation from focal consideration of consciousness,
affect and social factors, two fronts have been opened on this
classical position.

The first front is well-publicised and highly visible. Both Searle
(1992) and Edelman (1992) refuse to grant any special status to
information-processing in explanation of mental process. In contrast,
they argue, we should focus on Neuroscience on the one hand and
Consciousness on the other. The other front is ultimately the more
compelling one. It consists of those researchers from inside CS who
are currently working on consciousness, affect and social factors and
do not see any incompatibility between this research and their vision
of CS, which is that of a Science of Mind (see Dennett 1993, O
Nuallain (in press) and Mc Kevitt and Partridge 1991, Mc Kevitt and Guo
1994).

References

Dennett, D. (1993) Review of John Searle's "The Rediscovery of the
Mind". The Journal of Philosophy 1993, pp 193-205

Edelman, G.(1992) Bright Air, Brilliant Fire. Basic Books

Mc Kevitt, P. and D. Partridge (1991) Problem description and
hypothesis testing in Artificial Intelligence In ``Artificial
Intelligence and Cognitive Science '90'', Springer-Verlag British
Computer Society Workshop Series, McTear, Michael and Norman Creaney
(Eds.), 26-47, Berlin, Heidelberg: Springer-Verlag.  Also, in
Proceedings of the Third Irish Conference on Artificial Intelligence
and Cognitive Science (AI/CS-90), University of Ulster at Jordanstown,
Northern Ireland, EU, September and as Technical Report 224,
Department of Computer Science, University of Exeter, GB- EX4 4PT,
Exeter, England, EU, September, 1991.

Mc Kevitt, P. and Guo, Cheng-ming (1995) From Chinese rooms to Irish
rooms: new words on visions for language. Artificial Intelligence
Review Vol. 8. Dordrecht, The Netherlands: Kluwer-Academic Publishers.
(unabridged version) First published: International Workshop on
Directions of Lexical Research, August, 1994, Beijing, China.

O Nuallain, S (in press) The Search for Mind: a new foundation for
CS. Norwood: Ablex

Pylyshyn, Z.(1984) Computation and Cognition. MIT Press

Searle, J (1992) The rediscovery of the mind. MIT Press.

Von Eckardt, B. (1993) What is Cognitive Science? MIT Press

WORKSHOP TOPICS:

The tension which riddles current CS can therefore be stated thus: CS,
which gained its initial capital by adopting the computational
metaphor, is being constrained by this metaphor as it attempts to
become an encompassing Science of Mind. Papers are invited for this
workshop which:


* Address the central tension

* Propose an overall framework for CS (as attempted, inter alia,
  by O Nuallain (in press))

* Explicate the relations between the disciplines which comprise CS.

* Relate educational experiences in the field

* Describe research outside the framework of classical
  computationalist CS in the context of an alternative framework

* Promote a single logico-mathematical formalism as a theory of
  Mind (as attempted by Harmony theory)

* Disagree with the premise of the workshop


Other relevant topics include:

* Classical vs. neuroscience representations

* Consciousness vs. Non-consciousness

* Dictated vs. emergent behaviour

* A life/Computational intelligence/Genetic algorithms/Connectionism

* Holism and the move towards Zen integration


The workshop will focus on three themes:

* What is the domain of Cognitive Science ?

* Classic computationalism and its limitations

* Neuroscience and Consciousness


WORKSHOP FORMAT:

Our intention is to have as much discussion as possible during the
workshop and to stress panel sessions and discussion rather than
having formal paper presentations.  The workshop will consist of
half-hour presentations, with 15 minutes for discussion at the end of
each presentation and other discussion sessions.  A plenary session at
the end will attempt to resolve the themes emerging from the different
sessions.

ATTENDANCE:

We hope to have an attendance between 25-50 people at the workshop.

Given the urgency of the topic, we expect it to be of interest not
only to scientists in the AI/Cognitive Science (CS) area, but also to
those in other of the sciences of mind who are curious about CS.
We envisage researchers from Edinburgh, Leeds, York, Sheffield and Sussex
attending from within England and many overseas visitors as the
Conference Programme is looking very international.


SUBMISSION REQUIREMENTS:

Papers of not more than 8 pages should be submitted by electronic mail
(preferably uuencoded compressed postscript) to Sean O Nuallain at the
E-mail address(es) given below.  If you cannot submit your paper by
E-mail please submit three copies by snail mail.

*******Submission Deadline:  February 13th 1995
*******Notification Date:    February 25th 1995
*******Camera ready Copy:    March 10th 1995


PUBLICATION:

Workshop notes/preprints will be published.  If there is sufficient
interest we will publish a book on the workshop possibly with the
American Artificial Intelligence Association (AAAI) Press.

WORKSHOP CHAIR:

Sean O Nuallain  
((Before Dec 23:))
Knowledge Systems Lab,                       
Institute for Information Technology,
National Research Council,
Montreal Road, Ottawa
Canada K1A OR6

Phone:  1-613-990-0113
E-mail: sean at ai.iit.nrc.ca
FaX:    1-613-95271521

((After Dec 23:))
Dublin City University,
IRL- Dublin 9, Dublin
Ireland, EU

WWW:   http://www.compapp.dcu.ie
Ftp:   ftp.vax1.dcu.ie
E-mail: onuallains at dcu.ie
FaX: 353-1-7045442
Phone: 353-1-7045237


AISB-95 WORKSHOPS AND TUTORIALS CHAIR:

    Dr. Robert Gaizauskas
    Department of Computer Science
    University of Sheffield
    211 Portobello Street
    Regent Court
    Sheffield S1 4DP
    U.K.  E-mail:  robertg at dcs.shef.ac.uk
    WWW:     http://www.dcs.shef.ac.uk/
    WWW:     http://www.shef.ac.uk/
    Ftp:     ftp.dcs.shef.ac.uk
    FaX:     +44 (0) 114 278-0972
    Phone:   +44 (0) 114 278-5572


AISB-95 CONFERENCE/LOCAL ORGANISATION CHAIR:

Paul Mc Kevitt
Department of Computer Science
Regent Court
211 Portobello Street
University of Sheffield
GB- S1 4DP, Sheffield
England, UK, EU.

E-mail:    p.mckevitt at dcs.shef.ac.uk
WWW:       http://www.dcs.shef.ac.uk/
WWW:       http://www.shef.ac.uk/
Ftp:       ftp.dcs.shef.ac.uk
FaX:       +44 (0) 114-278-0972
Phone:     +44 (0) 114-282-5572 (Office)   
                       282-5596 (Lab.)    
          	       282-5590 (Secretary)


AISB-95 REGISTRATION:

        Alison White
        AISB Executive Office
        Cognitive and Computing Sciences (COGS)
        University of Sussex
        Falmer, Brighton
        England, UK, BN1 9QH

        Email:	alisonw at cogs.susx.ac.uk
        WWW: http://www.cogs.susx.ac.uk/users/christ/aisb
        Ftp: ftp.cogs.susx.ac.uk/pub/aisb
        Tel:	+44 (0) 1273 678448
        Fax:	+44 (0) 1273 671320


AISB-95 ENQUIRIES:

        Debbie Daly,
        Administrative Assistant, AISB-95,
	Department of Computer Science,
        Regent Court,
        211 Portobello Street,
	University of Sheffield,
	GB- S1 4DP, Sheffield,
        UK, EU.
	
	Email:    debbie at dcs.shef.ac.uk (personal communication)
	Fax:      +44 (0) 114-278-0972
	Phone:    +44 (0) 114-278-5565                (personal)
                                 -5590                (messages)

	Email:    aisb95 at dcs.shef.ac.uk                    (for auto responses)
        WWW: http://www.dcs.shef.ac.uk/aisb95      [Sheffield Computer Science]
        Ftp: ftp.dcs.shef.ac.uk (cd aisb95)
        WWW: http://www.shef.ac.uk/              [Sheffield Computing Services]
        Ftp: ftp.shef.ac.uk (cd aisb95)
        WWW: http://ijcai.org/)  (Email welty at ijcai.org)   [IJCAI-95, MONTREAL]
        WWW: http://www.cogs.susx.ac.uk/users/christ/aisb [AISB SOCIETY SUSSEX]
        Ftp: ftp.cogs.susx.ac.uk/pub/aisb 


VENUE:

The venue for registration and all conference events is: 

	Halifax Hall of Residence,
	Endcliffe Vale Road,
	GB- S10 5DF, Sheffield,
        UK, EU.

	FaX: +44 (0) 114-268-4227	
	Tel: +44 (0) 114-268-2758 (24 hour porter)
	Tel: +44 (0) 114-266-4196 (manager)


SHEFFIELD:

Sheffield is one of the friendliest cities in Britain and is situated
well having the best and closest surrounding countryside of any major
city in the UK. The Peak District National Park is only minutes
away. It is a good city for walkers, runners, and climbers.  It has
two theatres, the Crucible and Lyceum. The Lyceum, a beautiful
Victorian theatre, has recently been renovated.  Also, the city has
three 10 screen cinemas.  There is a library theatre which shows more
artistic films. The city has a large number of museums many of which
demonstrate Sheffield's industrial past, and there are a number of
Galleries in the City, including the Mapping Gallery and Ruskin.  A
number of important ancient houses are close to Sheffield such as
Chatsworth House. The Peak District National Park is a beautiful site
for visiting and rambling upon.  There are large shopping areas in the
City and by 1995 Sheffield will be served by a 'supertram' system: the
line to the Meadowhall shopping and leisure complex is already open.

The University of Sheffield's Halls of Residence are situated on the
western side of the city in a leafy residential area described by John
Betjeman as ``the prettiest suburb in England''.  Halifax Hall is
centred on a local Steel Baron's house, dating back to 1830 and set in
extensive grounds. It was acquired by the University in 1830 and
converted into a Hall of Residence for women with the addition of a
new wing.


ARTIFICIAL INTELLIGENCE AT SHEFFIELD:

Sheffield Computer Science Department has a strong programme in
Cognitive Systems and is part of the University's Institute for
Language, Speech and Hearing (ILASH).  ILASH has its own machines and
support staff, and academic staff attached to it from nine
departments. Sheffield Psychology Department has the Artificial
Intelligence Vision Research Unit (AIVRU) which was founded in 1984 to
coordinate a large industry/university Alvey research consortium
working on the development of computer vision systems for autonomous
vehicles and robot workstations.  Sheffield Philosophy Department has
the Hang Seng Centre for Cognitive Studies, founded in 1992, which
runs a workshop/conference series on a two-year cycle on topics of
interdisciplinary interest. (1992-4: 'Theory of mind'; 1994- 6:
'Language and thought'.)






From lautrup at connect.nbi.dk  Fri Dec 16 12:01:20 1994
From: lautrup at connect.nbi.dk (Benny Lautrup)
Date: Fri, 16 Dec 94 12:01:20 MET
Subject: Paper available: "Massive Weight Sharing: A Cure for ..."
Message-ID: <mailman.722.1149591322.29955.connectionists@cs.cmu.edu>

FTP-host: connect.nbi.dk
FTP-file: neuroprose/lautrup.massive.ps.Z

----------------------------------------------

The following paper is now available:

Massive Weight Sharing: A Cure for Extremely Ill-posed Learning [12 pages]

B. Lautrup, L.K. Hansen, I. Law, N. Moerch, C. Svarer and S.C. Strother
CONNECT, The Niels Bohr Institute, University of Copenhagen, Denmark

Presented at the workshop on Supercomputing in Brain Research: From Tomography
to Neural Networks, HLRZ, Juelich, Germany, November 21-23, 1994

Abstract:
In most learning problems, adaptation to given examples is  well-posed  because
the number of examples far exceeds the number of  internal  parameters  in  the
learning machine. Extremely ill-posed learning problems are, however, common in
image and spectral analysis. They are characterized by a vast number of  highly
correlated inputs, e.g. pixel or pin values, and a modest number  of  patterns,
e.g. images or spectra. In this paper we show, for the case of  a  set  of  PET
images differing only in the values of  one  stimulus  parameter,  that  it  is
possible to train a neural network to learn the underlying rule  without  using
an excessive number of network weights or large amounts of computer  time.  The
method is based upon the observation that the standard learning rules  conserve
the subspace spanned by the input images.

Please do not reply directly to this message.

-----------------------------------------------

FTP-instructions:

unix> ftp connect.nbi.dk (or 130.225.212.30)
ftp> Name: anonymous
ftp> Password: your e-mail address
ftp> cd neuroprose
ftp> binary
ftp> get lautrup.massive.ps.Z
ftp> quit
unix> uncompress lautrup.massive.ps.Z

-----------------------------------------------

Benny Lautrup,
Computational Neural Network Center (CONNECT)
Niels Bohr Institute
Blegdamsvej 17
2100 Copenhagen
Denmark

Telephone: +45-3532-5200
Direct:    +45-3532-5358
Fax:       +45-3142-1016
e-mail:    lautrup at connect.nbi.dk


From jaap.murre at mrc-apu.cam.ac.uk  Fri Dec 16 08:46:00 1994
From: jaap.murre at mrc-apu.cam.ac.uk (Jaap Murre)
Date: Fri, 16 Dec 94 13:46:00 GMT
Subject: Papers and PC demo available
Message-ID: <9412161346.AA00917@rigel.mrc-apu.cam.ac.uk>


The following three files have recently (15-12-1994) been added to our
ftp site (ftp://ftp.mrc-apu.cam.ac.uk/pub/nn/murre):

File 1:
nnga1.ps   Happel, B.L.M., & J.M.J. Murre (1994). Design and
           evolution of modular neural network architectures. Neural
           Networks, 7, 985-1004. (About 0.5 Mb; ps.Z version is
           recommended.)

Abstract:  To investigate the relations between structure and function in
both artificial and natural neural networks, we present a series of
simulations and analyses with modular neural networks. We suggest a
number of design principles in the form of explicit ways in which neural
modules can cooperate in recognition tasks. These results may
supplement recent accounts of the relation between structure and
function in the brain. The networks used consist out of several modules,
standard subnetworks that serve as higher-order units with a distinct
structure and function. The simulations rely on a particular network
module called CALM (Murre, Phaf, and Wolters, 1989, 1992). This
module, developed mainly for unsupervised categorization and learning,
is able to adjust its local learning dynamics. The way in which modules
are interconnected is an important determinant of the learning and
categorization behaviour of the network as a whole. Based on arguments
derived from neuroscience, psychology, computational learning theory,
and hardware implementation, a framework for the design of such
modular networks is laid-out. A number of small-scale simulation
studies shows how intermodule connectivity patterns implement 'neural
assemblies' (Hebb, 1949) that induce a particular category structure in
the network. Learning and categorization improves as the induced
categories are more compatible with the structure of the task domain. In
addition to structural compatibility, two other principles of design are
proposed that underlie information processing in interactive activation
networks: replication and recurrence. 
      Because a general theory for relating network architectures to
specific neural functions does not exist, we extend the biological
metaphor of neural networks, by applying genetic algorithms (a
biocomputing method for search and optimization based on natural
selection and evolution) to search for optimal modular network
architectures for learning a visual categorization task. The best
performing network architectures seemed to have reproduced some of
the overall characteristics of the natural visual system, such as the
organization of coarse and fine processing of stimuli in separate
pathways. A potentially important result is that a genetically defined
initial architecture cannot only enhance learning and recognition
performance, but it can also induce a system to better generalize its
learned behaviour to instances never encountered before. This may
explain why for many vital learning tasks in organisms only a minimal
exposure to relevant stimuli is necessary.

File 2:
chaos1.ps  Happel, B.L.M., & J.M.J. Murre (submitted). Evolving
           complex dynamics in modular interactive neural networks.
           Submitted to Neural Networks. (This is a large file: 1.5 Mb!
           Retrieve the ps.Z version if possible.)

Abstract:  Computational simulation studies, carried out within a
general framework of modular neural network design, demonstrate that
networks consisting of many interacting modules provide a variety of
different neural processing principles. The dynamics underlying these
behaviors range from simple linear separation of input vectors in
individual modules, to oscillations, evoking chaotic regimes in the
activity evolution of a network. As opposed to static representations in
conventional neural network models, information in oscillatory networks
is represented as space-time patterns of activity. Chaos in a neural
network can serve as: (i) a novelty filter, (ii) explorative deterministic
noise, (iii) a fundamental form of neural activity that provides
continuous, sequential access to memory patterns, and (iv) a mechanism
that underlies the formation of complex categories. An experiment in the
artificial evolution of modular neural architectures, demonstrates that by
manipulating modular topology and parameters governing local learning
and activation processes, "genetic algorithms" can effectively explore
complex interactive dynamics to construct efficient, modular neural
architectures for pattern categorization tasks. A particularly striking
result is that coupled, oscillatory circuits were installed by the genetic
algorithm, inducing the formation of fractal category boundaries.
Dynamic representations in these networks, can significantly reduce
sequential interference due to overlapping static representations in
learning neural networks.

File 3:
The above two papers, among others, describe a digit recognition
network. A demonstration version of this can be retrieved for PCs (486
DX recommended):

digidemo.zip           (unzip with PKUNZIP; contains several files)

Some documentation is included with the program. One of its features is
retraining on digits that go wrong. Only the wrong digits are retrained,
without catastrophic interference with the other digits.

With questions and remarks, contact either Bart Happel at Leiden
University (happel at rulfsw.leiden.univ) or Jaap Murre at the MRC
Applied Psychology Unit (jaap.murre at mrc-apu.cam.ac.uk).




From LINDBLAD at vana.physto.se  Fri Dec 16 08:44:45 1994
From: LINDBLAD at vana.physto.se (LINDBLAD@vana.physto.se)
Date: Fri, 16 Dec 1994 15:44:45 +0200
Subject: IBM ZISC Paper
Message-ID: <01HKPT2KNC4Y8Y6ZY4@vana.physto.se>

FTP-HOST: archive.cis.ohio-state.edu
FTP-FILE: pub/neuroprose/eide.zisc.ps.Z

        The file eide.zisc.ps. has been placed into the 
Neuroprose repository:
  
"An implementation of the Zero Instruction Set Computer (ZISC036) 
on a PC/ISA-bus card" (14 pages)
A. Eide, Th. Lindblad, C.S. Lindsey, M. Minerskjoeld, G.Sekhniaidze
and G. Szekely

Abstract: The new IBM Zero Instruction Set Computer (ZISC036) chip has
been implemented on a PC/ISA-bus card. The chip has 36 RBF-like neurons.
It is highy parallel and cascadable with on-chip learning. A card with
two ZISC036 chips was built and tested with a noisy character recognition
"benchmark".
 


From LINDBLAD at vana.physto.se  Fri Dec 16 08:46:00 1994
From: LINDBLAD at vana.physto.se (LINDBLAD@vana.physto.se)
Date: Fri, 16 Dec 1994 15:46:00 +0200
Subject: WWW Page on Neural Nets in High Energy Physics
Message-ID: <01HKPT45B9N68Y6ZY4@vana.physto.se>

 	We would like to announce the installation of
the WWW homepage "Neural Networks in High Energy Physics". 
The address is:

       "http://www1.cern.ch/NeuralNets/nnwInHep.html"

	Both hardware and software neural network techniques have
been used in high energy physics. Hardware neural networks are
used for real-time data selection, while software neural networks
are used in off-line analysis to enhance signal to background
discrimination. We include a long reference list of work done 
in these areas, as well as news of recent developments. 
	Of most general interest is the extensive page on 
commercial hardware neural networks. Descriptions are given
of VLSI chips (e.g. the new IBM ZISC chip with RBF neurons),
accelerator boards, and neurocomputers.
                Clark S. Lindsey (lindsey at msia02.msi.se)
                Bruce Denby (denby at pisa.infn.it)
                Thomas Lindblad (lindblad at vana.physto.se)


From ricart at picard.jmb.bah.com  Fri Dec 16 10:32:22 1994
From: ricart at picard.jmb.bah.com (Rick Ricart)
Date: Fri, 16 Dec 94 10:32:22 EST
Subject: Apology for Flame, and Survey
Message-ID: <9412161532.AA03333@picard.jmb.bah.com>


I agree with John Pollock that reviewing fees should not be instituted.
I'm not sure what prompted the unprecedented decision by WCNN, but if one 
reason for the fee is to increase the overall quality of the submissions, 
I offer the following alternative.  Have each society (IEEE, INNS, etc.) 
establish and publish minimum acceptance criteria for papers.  These criteria 
might include, for example, precise algorithm and parameter descriptions so that 
experiments can be reproduced by others.  I know this is a novel idea for 
some current scientific and engineering societies and publishers, but I think
we're ready for it.

The real reason for the reviewing fee might be, of course, financial.  All 
reviewers are very busy individuals with many professional duties.  Some, as is
evident, feel they should be reimbursed for "volunteering" their precious
time for the given society's benefit.  My answer to the problem is, "don't 
volunteer."  There are many other professionals in the field that would 
gladly volunteer their time to review papers given the opportunity; especially
if they have clear minimum acceptance guidelines available from the given
society.

These are my personal feelings and in no way do they represent an official 
Booz-Allen & Hamilton, Inc view point.

Rick Ricart					Associate
Booz-Allen & Hamilton   Advanced Computational Technologies Practice
McLean, VA 22020			Phone: (703) 902-5494
email: ricart at picard.jmb.bah.com


From timxb at faline.bellcore.com  Fri Dec 16 11:26:07 1994
From: timxb at faline.bellcore.com (Timothy X Brown)
Date: Fri, 16 Dec 1994 11:26:07 -0500
Subject: Apology for Flame, and Survey
Message-ID: <199412161626.LAA21111@faline.bellcore.com>

Review fees to cover costs may or may not be justified, but it is a
dangerous precedent, especially when an implicit goal was to limit
entry. Organizers of conferences/journals/grants/positions can only
lose by restricting the submission that they receive. Each of the
"problems" Jordan Pollack mentions can be addressed in slightly
different ways:

>a) filtering out computer-generated articles,
I thought "filtering out" is the quintissential role of the review
process. By having some standards for acceptance, such fluff can be
removed. One of the roles of a conference is to do some of this before
hand so I don't waste valuable time on low information content.

>b) authors not showing up to deliver talks or posters,
Have no review fee, but the authors must submit the conference fee
which is then refundable if the paper is not accepted.

>c) expense of supporting travel costs for both plenary 
   speakers and conference organizers, and
If the organizers cannot organize a cost effective conference, get
someone else to do it. I know that the American Social Sciences
Association runs a huge 28 parallel session conference, and conference
fees last year (this was in downtown Boston) were $35 with a $15
student rate! Of course, banquets (choose from one of 10), etc, were
extra and water was the only refreshment provided, but what is the
point of the conference? BTW, I've organized conferences without my
expenses being covered.

>d) declining membership/subscription revenues.
Orgainize smaller conferences (with <6 parallel sessions), smaller
print runs, etc. If profits are the only motive, put a centerfold in
the magazine and have naked dancers at the banquet.

	Tim

Timothy X Brown
MRE 2E-378
Adaptive Systems Research
Bell Communications Research
445 South Street
Morristown, NJ 07960
Tel: (201) 829-4314
Fax: (201) 829-5888



From singh at psyche.mit.edu  Fri Dec 16 16:13:43 1994
From: singh at psyche.mit.edu (Satinder Singh)
Date: Fri, 16 Dec 94 16:13:43 EST
Subject: Transfer/Interference in control problems...
Message-ID: <9412162113.AA17889@psyche.mit.edu>

	This is another note on the topic of transfer and interference
across problems. 

	I have investigated this issue *a bit* in the reinforcement
learning setting of an agent required to solve a SET of hierarchically
structured control problems in the *same* environment. Reinforcement
learning (RL) algorithms solve control problems by using control
experience to update utility functions. With experience, the utility
function improves and therefore so does the resulting ``greedy''
decision strategy.

	I studied the case of compositionally structured control
tasks, where complex tasks are sequences of simpler tasks. I showed
how a modular RL architecture (that is based on Jacobs, Jordan, Nowlan
and Hinton's mixture of experts architecture) is able to reuse utility
functions learned for simple tasks to quickly construct utility
functions for more complex tasks in the hierarchy. The composition of
a complex task is not available to the agent, but has to be learned. 
The use of ``shaping'' to encourage transfer is also illustrated. 

	Anonymous ftp instructions for the above work follow -- the
filename is singh-MLJ-1.ps.Z (another file of possible interest is
singh-ML92.ps.Z).

	Several other RL researchers have since worked on this topic.
See Dayan and Hinton's paper on ``Feudal Reinforcement Learning'' in
NIPS-5 for more recent work and uptodate references.
================================================================

unix> ftp envy.cs.umass.edu

Name: anonymous
Password: [your ID]
ftp> cd pub/singh
ftp> binary
ftp> get <filename>.ps.Z
ftp> bye

unix> uncompress <filename>.ps.Z
unix> [your command to print PostScript file] <filename>.ps


From N.Sharkey at dcs.shef.ac.uk  Fri Dec 16 12:06:27 1994
From: N.Sharkey at dcs.shef.ac.uk (N.Sharkey@dcs.shef.ac.uk)
Date: Fri, 16 Dec 94 17:06:27 GMT
Subject: About sequential learning (or interference)
In-Reply-To: Stephen Hanson's message of Wed, 14 Dec 1994 15:20:45 -0500 (EST) <YivpCRu1GEMnA0vvJt@tractatus.siemens.com>
Message-ID: <9412161706.AA01599@entropy.dcs.shef.ac.uk>


>of course avoiding "interference" is
>another way of preventing generalization.

>As usual there is a tradeoff here.

Yes there is a trade-off but not with interference as Hetherington has pointed
out.

In fact, with backprop, the way to entirely eliminate interference is
to get a good approximation to the total underlying function that is
being sampled. For example, with an autoencoder memory, if the there
is good extraction of the identity function then there will be no
interference from training on sucessive memory sets (and of course
little need for further training).

The trade-off is between old-new discrimination and generalisation. Definitionally,
as one improves the other collapses.

In the paper cited by Heterington (which is now under journal
submission). We present a formally guaranteed solution to the
interference and discrimination problem (the HARM) model, but it
demands exponentially increasing computational resources. It is really
used to show the problems of other localisation solutions (French,
Murre, Kruschke etc.).

We also report some interesting empirical (simulation) results of this trade-off in a
much shorter paper:

Sharkey, NE, and Sharkey, AJC, (in press) Interference and
Discrimination in Neural Net Memory. In Joe Levy, Dimitrios
Bairaktaris, John Bullinaria and Paul Cairns.(Ed) Connectionist Models
of Memory and Language, UCL press.

If anyone is interested I will mail a postscript copy of the tech report to them.

Sharkey, N.E., & Sharkey, A.J.C. Understanding Catastrophic
  Interference in neural Nets.  Technical Report, Department of Computer
  Science, University of Sheffield, U.K. 

Abstract

A number of recent simulation studies have shown that when feedforward
neural nets are trained, using backpropagation, to memorize sets of
items in sequential blocks and without negative exemplars, severe
retroactive interference or {\em catastrophic forgetting} results.
Both formal analysis and simulation studies are employed here to show
why and under what circumstances such retroactive interference arises.
The conclusion is that, on the one hand, approximations to "ideal"
network geometries can entirely alleviate interference, but at the
cost of a breakdown in discrimination between input patterns that have
been learned and those that have not: {\em catastrophic remembering}.
On the other hand, localized geometries for subfunctions eliminate the
discrimination problem but are easily disrupted by new training sets
and thus cause {\em catastrophic interference}.  The paper concludes
with a Hebbian Autoassociative Recognition Memory (HARM) model which
provides a formally guaranteed solution to the problems of
interference and discrimination. This is then used as a yardstick with
which to evaluate other proposed solutions.

noel

 Noel Sharkey  	   	   
 Professor of Computer Science   
 Department of Computer Science  
 Regent Court                    
 University of Sheffield 	 
 S1 4DP, Sheffield, UK           

 N.Sharkey at dcs.shef.ac.uk

FAX: (0742) 780972



From dyyeung at cs.ust.hk  Sat Dec 17 09:54:27 1994
From: dyyeung at cs.ust.hk (Dr. D.Y. Yeung)
Date: Sat, 17 Dec 94 09:54:27 HKT
Subject: elitism at NIPS
Message-ID: <9412170154.AA17552@cs.ust.hk>

> The folks involved in running NIPS are aware that in the past, some people
> have felt the conference was biased against outsiders.  As the program
> chair for NIPS*94, I want to mention some steps we took to reduce that
> perception:
> 
> * This year we required authors to submit full papers, not just abstracts.
>   That made it harder for "famous" people to get by on their reputation
>   alone, and easier for "non-famous" people to get good papers in.
> 
> * This year we required reviewers to provide real feedback to all authors.
>   In order to make this less painful, we completely redesigned the review
>   form (it's publicy accessible via the NIPS home page) and, for the first
>   time, accepted reviews by email.  Everyone liked getting comments from
>   the reviewers, and authors whose papers were not accepted understood why.
> 
> * We continue to recruit new people for positions in both the organizing
>   and program committees.  It's not the same dozen people year after year.
>   We also have a large reviewer pool:  176 people served as reviewers
>   this year.
> 
> * We tend to bring in "outsiders" as our invited speakers, rather than
>   the usual good old boys.  This year's invited speakers included Francis
>   Crick of the Salk Insititute, Bill Newsome (a neuroscientist from 
>   Stanford), and Malcolm Slaney (a signal processing expert formerly with
>   Apple and now at Interval Research.)  None had been to NIPS before.

Another step that the organizers of NIPS may consider is to have a
"blind" review process, which is an even more positive step to reduce
that perception.  This practice has been used in some other good
conferences too.

Regards,

Dit-Yan Yeung


From raymond at fit.qut.edu.au  Sat Dec 17 20:24:16 1994
From: raymond at fit.qut.edu.au (Raymond Lister)
Date: Sat, 17 Dec 94 20:24:16 EST
Subject: elitism at NIPS
Message-ID: <199412171024.UAA20868@fitmail.fit.qut.edu.au>


> From Dave_Touretzky at cs.cmu.edu Sat Dec 17 14:53:02 1994
> To: Connectionists at cs.cmu.edu
> Subject: elitism at NIPS
> Date: Fri, 16 Dec 94 03:22:31 EST
> 
> ...
> 
> * This year we required authors to submit full papers, not just abstracts.
>   That made it harder for "famous" people to get by on their reputation
>   alone, and easier for "non-famous" people to get good papers in.

The move to full papers was a good change.  It would be better still if the
copies of papers sent to referees did not contain the names and addresses
of authors.

Raymond Lister,
School of Computing Science,
Queensland University of Technology,
AUSTRALIA

Internet: raymond at fitmail.fit.qut.edu.au


From gluck at pavlov.rutgers.edu  Sat Dec 17 10:45:25 1994
From: gluck at pavlov.rutgers.edu (Mark Gluck)
Date: Sat, 17 Dec 94 10:45:25 EST
Subject: Rutgers Grad Program in BEHAVIORAL AND NEURAL SCIENCES
Message-ID: <9412171545.AA01777@james.rutgers.edu>

---------------------------------------------------------------------
 
       Seeking Applications for Fall 1995 for Ph.D. Program in

		    BEHAVIORAL AND NEURAL SCIENCES

		      Rutgers University, Newark

	   Target date for applications is JANUARY 20, 1995

---------------------------------------------------------------------

If you are considering graduate study in Cognitive, Integrative,
Molecular, or Computational Neuroscience, you may be interested in
Rutgers' new interdisciplinary research-oriented graduate program in
Behavioral and Neural Sciences (BNS).

The BNS aims to provide students with a rigorous understanding of the
basic tenets and underpinnings of modern neuroscience.  The program
emphasizes the multidisciplinary nature of this endeavor, and offers
specific research training in Behavioral and Cognitive Neuroscience
and Molecular, Cellular and Systems Neuroscience.  These research
areas represent different but complementary approaches to contemporary
issues in behavioral and molecular neuroscience and can emphasize
either human or animal studies.

The graduate program is offered by two distinct university units: the
newly established Center for Molecular and Behavioral Neuroscience
(CMBN) and the Institute of Animal Behavior (IAB). These two units
work together but each has its own special emphasis.

Research at the CMBN emphasizes integration across levels of analysis
and traditional disciplinary boundaries. The CMBN is one of the
leading places in this country for the study of the neural bases of
behavior and cognition in humans and other animals. Behavioral
research areas include the study of memory, language (both signed and
spoken), motor-control, and vision. Clinically relevant research areas
are the study of the physiological and pharmacological aspects of
schizophrenia, epilepsy and Parkinson's disease and molecular genetics
of reading disorders and well as neuroendocrinology. We have a
computational program for students interested in pursuing
neural-network models as a tool for understanding psychological and
biological issues.  There is also a strong focus on single cell (patch
clamp, intracellular and extracellular) electrophysiology and
multi-unit recording, systems analysis, neuroanatomy and in vivo
microdialysis.

The IAB offers a unified program in psychobiology and ethological
patterns of behavior, with an emphasis on evolution, development and
reproduction, as well as the neurogenesis and recovery of function
from brain damage.


Other Info 
---------- 
At present the CMBN supports up to 40 students with 12-month renewable
assistantships for a period of four years. The curent stipend for
first year students is $12,750; this includes tuition remission and
excellent healthcare benefits. The IAB supported students receive D.S.
Lehrman Fellowships which include a 12-month stipend of approximately
$10,500 for four years and tuition remission.  In addition, the
Johnson & Johnson pharmaceutical company's Foundation has provided
four Excellence Awards which increase students' stipends by $5,000.
Several other fellowships are offered.  More information is available
in our graduate brochure.

The Rutgers-Newark campus (as distinct from the New Brunswick campus),
is 30 minutes outside New York City, and close to other major
university research centers at NYU, Columbia, and Princeton, as well
as major industrial research labs in Northern NJ, including ATT,
Bellcore, Siemens, and NEC.


Faculty Associated With Rutgers Behavioral & Neural Sciences Ph.D. Program
--------------------------------------------------------------------------
	FACULTY - RUTGERS

Elizabeth Abercrombie (Ph.D., Princeton), neurotransmitters and behavior [CMBN]
Colin Beer (Ph.D., Oxford), ethology [IAB]
April Benasich (Ph.D., New York), infant perception and cognition [CMBN]
Ed Bonder (Ph.D., Pennsylvania), cell biology [Biology]
Linda Brzustowicz (M.D.,Ph.D., Columbia), human genetics [CMBN]
Gyorgy Buzsaki (Ph.D., Budapest), systems neuroscience [CMBN]
Mei-Fang Cheng (Ph.D., Bryn Mawr) neuroethology/neurobiology [IAB]
Ian Creese (Ph.D., Cambridge), neuropsychopharmacology [CMBN]
Doina Ganea (Ph.D., Illinois Medical School), molecular immunology [Biology]
Alan Gilchrist (Ph.D., Rutgers), visual perception [Psychology]
Mark Gluck (Ph.D.,Stanford), learning, memory and neural computation [CMBN]
Ron Hart (Ph.D., Michigan), molecular neuroscience [Biology]
G. Miller Jonakait (Ph.D., Cornell Medical College), neuroimmunology [Biology]
Judy Kegl (Ph.D., M.I.T.), linguistics/neurolinguistics [CMBN]
Barry Komisaruk (Ph.D., Rutgers), behavioral neurophysiology/pharmacology [IAB]
Sarah Lenington (Ph.D., Chicago), genetic basis of mating preference [IAB]
Joan Morrell (Ph.D., Rochester), cellular neuroendocrinology [CMBN]
Teresa Perney (Ph.D., Chicago), ion channel gene expression and function [CMBN]
Howard Poizner (Ph.D., Northeastern), language and motor behavior [CMBN]
Jay Rosenblatt (Ph.D., New York), maternal behavior [IAB]
Anne Sereno (Ph.D., Harvard), attention and visual perception [CMBN]
Maggie Shiffrar (Ph.D., Stanford), vision and motion perception[CMBN]
Harold Siegel (Ph.D., Rutgers) neuroendocrine mechanisms [IAB]
Ralph Siegel (Ph.D., McGill), neuropsychology of visual perception [CMBN]
Donald Stein (Ph.D., Oregon), neural plasticity [IAB]
Jennifer Swann (Ph.D., Michigan), neuroendocrinology [Biology]
Paula Tallal (Ph.D., Cambridge), neural basis of language development [CMBN]
James Tepper (Ph.D., Colorado), basal ganglia neurophysiology and anatomy [CMBN]
Beverly Whipple (Ph.D., Rutgers), women's health [Nursing]
Laszlo Zaborszky (Ph.D., Hungarian Academy), neuroanatomy of forebrain [CMBN]

	BNS FACULTY - UMDNJ 
Barry Levin (M.D., Emory Medical) neurobiology
Benjamin Natelson (M.D., Pennsylvania) stress and distress
Allan Siegel (Ph.D., SUNY-Buffalo), aggressive behavior
Walter Tapp (Ph.D., Cornell), primate models of cognitive function
	
        ASSOCIATES OF CMBN
Izrail Gelfand (Ph.D., Moscow State), biology of cells [Biology]
Richard Katz (Ph.D., Bryn Mawr), psychopharmacology [Ciba Geigy]
David Tank (Ph.D., Cornell), neural plasticity [Bell Labs]


For More Information or an Application
--------------------------------------
If you are interested in applying to our graduate program, or possibly
applying to one of the labs as a post-doc, research assistant or
programmer, please contact us via one of the following:

	Dr. Gyorgy Buzsaki or Dr. Mark A. Gluck 
        CMBN, Rutgers University
        197 University Ave.
        Newark, New Jersey  07102

	Phone (Secretary, Ann Kutyla):  (201) 648-1080 (Ext. 3200) 
        Fax:    			(201) 648-1272
	Email:  buzsaki at axon.rutgers.edu or
		gluck at pavlov.rutgers.edu

We will be happy to send you info on our research and graduate
program, as well as set up an a possible visit to the Neuroscience
Center here at Rutgers-Newark.


INTERNET INFORMATION:
---------------------
Additional information on this program can be obtained 
over the internet via World Wide Web at:

	http://www.cmbn.rutgers.edu/

Please be warned that it is still under construction.



From sef+ at cs.cmu.edu  Sat Dec 17 15:20:28 1994
From: sef+ at cs.cmu.edu (Scott E. Fahlman)
Date: Sat, 17 Dec 94 15:20:28 EST
Subject: elitism at NIPS 
In-Reply-To: Your message of Sat, 17 Dec 94 20:24:16 -0500.
             <199412171024.UAA20868@fitmail.fit.qut.edu.au> 
Message-ID: <mailman.723.1149591322.29955.connectionists@cs.cmu.edu>


Personally, I think that "blind" reviewing is a bad idea because it is
dishonest.  In a field like this one, it is easy in 90% of the cases for an
experienced reviewer to tell who is the author of a paper, or at least what
group the paper is from.  I think it's better to be explicit about the lack
of anonymity, and to challenge the reviewers to consciously try to rise
above any "in group" bias than to provide a show of anonymity that is
really a fraud in most cases.

I also believe that in some cases the identity of the author does provide
essential context to the reviewer.  If I were to present evidence that my
own Cascade-Correlation algorithm gets certain things wrong, a reviewer
might not have to worry as much about rookie mistakes in running Cascor as
he would in a paper from someone he has never heard of.  On the other hand,
if I claim that Cascor does certain things *better* than the competition,
the claim might warrant extra scrutiny because I am an interested party in
the debate.  Even more scrutiny is called for if I am known to have a
long-standing feud with the person whose work I am criticizing.

I know that this view is controversial.  Some of you will certainly argue
that such considerations have no place in science, and that every paper
must stand or fall on its content alone.  Perhaps that is true in a long
paper, but it may be impossible to present all the relevant context in an
eight-page paper. (Because of the NIPS style, this is more like a 5-page
paper published elsehwere.)

I don't know if this is the right forum for a debate on the merits of blind
reviewing.  I just thought it would be useful to point out that there are
some arguments on the other side.

-- Scott

===========================================================================
Scott E. Fahlman			Internet:  sef+ at cs.cmu.edu
Principal Research Scientist		Phone:     412 268-2575
School of Computer Science              Fax:       412 268-5576
Carnegie Mellon University		Latitude:  40:26:46 N
5000 Forbes Avenue			Longitude: 79:56:55 W
Pittsburgh, PA 15213			Mood:      :-)
===========================================================================


From peter at ai.iit.nrc.ca  Sun Dec 18 22:41:32 1994
From: peter at ai.iit.nrc.ca (Peter Turney)
Date: Mon, 19 Dec 1994 08:41:32 +0500
Subject: Workshop on Data Engineering for Inductive Learning
Message-ID: <9412191341.AA05163@ksl0j.iit.nrc.ca>




CALL FOR PARTICIPATION: Workshop on Data Engineering for Inductive Learning
---------------------------------------------------------------------------

IJCAI-95, Montreal (Canada), August 19/20/21, 1995


Objective
---------

In inductive learning, algorithms are applied to data. It is
well-understood that attention to both elements is critical -- unless
instances are represented so as to make its generalization methods
appropriate, no inductive learner can succeed. In applied work, it is
not uncommon for practitioners to spend the bulk of their time
exploring and transforming data in efforts to enable the use of
existing induction techniques.

Despite widespread acceptance of these facts, however, research reports
normally give data work short shrift. In fact, a report devoted mainly
to the data in an induction problem rather than to the algorithms that
process it might well be difficult to publish in mainstream machine
learning and neural network venues. 

Our goal in this workshop is to counterbalance the predominant focus on
algorithms by providing a forum in which data takes center stage.
Specifically, we invite discussion of issues relevant to data
engineering, which we define as the transformation of raw data into a
form useful as input to algorithms for inductive learning. Data
engineering is a concern in industrial and commercial applications of
machine learning, neural networks, genetic algorithms, and traditional
statistics. Among others, papers of the following kind are welcome:

1. Detailed case studies of data engineering in real-world applications 
   of inductive learning.

2. Descriptions of data engineering techniques or methods that have
   proven useful across a number of applications.

3. Studies of the data requirements of important inductive learning
   algorithms, the specifications to which data must be engineered for
   these algorithms to function.

4. Reports on software tools and environments for data engineering,
   including papers on "interactive induction" algorithms.

5. Empirical studies documenting the effect of data engineering on the
   success of induced models.

6. Surveys of data engineering practice in related fields: statistics,
   pattern recognition, etc. (but not problem-solving or 
   theorem-proving).

7. Papers on constructive induction, feature selection and related
   techniques.

8. Papers on (re)formulating a problem, to make it suitable
   for inductive learning techniques. For example, a paper
   on reformulating the problem of information filtering as
   learning to classify.

This workshop will enable an overview of current work in data
engineering. Since the problem of data engineering has received
relatively little published attention, it is difficult to anticipate
the work that will be presented at this workshop. We expect that the
workshop will make it possible to see common trends, shared problems,
and clever solutions that we cannot guess at, given our current,
limited view of data engineering. We have allowed ample time for
discussion of each paper (10 minutes), to foster an atmosphere
that will encourage data engineers to share their stories and to seek
common elements. We aim to leave the workshop with a vision of the
research directions that might bring science into data engineering.


Participation
-------------

During the workshop, we anticipate approximately 14 presentations. Each
paper will be given 25 minutes, 15 minutes for presentation and 10
minutes for discussion. There will be at most 30 participants in the
workshop. If you wish to participate in the workshop, you may either
submit a paper or a description of work that you have done (are doing,
plan to do) that is relevant to the workshop. Papers should be at most
10 pages long. The first page should include the title, the author's
name(s) and affiliation(s), a complete mailing address, phone number,
fax number, e-mail, an abstract of at most 300 words, and up to five
keywords. For those who do not choose to submit a paper, a description
of relevant work should be at most 1 page long and should include
complete address information. Workshop participants are required to
register for the main IJCAI-95 conference.

All submissions (papers or descriptions of relevant work) will be
reviewed by at least two members of the organizing committee. Please
send your submissions to the contact address below. Submissions should
be PostScript files, sent by e-mail. Accepted submissions will be
available before the workshop through ftp. Workshop participants will
also be given copies of the papers on the day of the workshop.

In selecting the papers, the committee will aim for breadth of coverage
of the topics listed above. Ideally, each of the eight kinds of papers
listed above would have at least one representative in the workshop.
A paper with new ideas on data engineering will be preferred to a high-
quality paper on a familiar idea.

The workshop organizers plan to publish revised versions of selected
papers from the workshop. The papers would be published either as a
book or as a special issue of a journal.

The exact date for the workshop has not yet been decided by IJCAI.
The workshop is one day in duration and will be held on one of
August 19, 20, or 21.


Schedule
--------

Deadline for submissions:		March 31, 1995
Notification of acceptance:		April 21, 1995
Submissions available by ftp:		April 28, 1995
Actual Workshop:			August 19/20/21, 1995


Organizing Committee
--------------------

Peter Turney, National Research Council (Canada)
Cullen Schaffer, CUNY/Hunter College (USA)
Rob Holte, University of Ottawa (Canada)


Contact Address
---------------

Dr. Peter Turney
Knowledge Systems Laboratory
Institute for Information Technology
National Research Council Canada
Ottawa, Ontario, Canada
K1A 0R6
(613) 993-8564 (office)
(613) 952-7151 (fax)
peter at ai.iit.nrc.ca









From ucganlb at ucl.ac.uk  Mon Dec 19 09:08:28 1994
From: ucganlb at ucl.ac.uk (Dr Neil Burgess - Anatomy UCL London)
Date: Mon, 19 Dec 94 14:08:28 +0000
Subject: catastrophic interference of BP
Message-ID: <187786.9412191408@link-1.ts.bcc.ac.uk>

Studies of catastrophic interference in BP networks are
interesting when considering such a network as a model of some human
(or animal) memory system. 
Is there any reason for doing that?
Neil


From ted at SPENCER.CTAN.YALE.EDU  Mon Dec 19 09:52:27 1994
From: ted at SPENCER.CTAN.YALE.EDU (ted@SPENCER.CTAN.YALE.EDU)
Date: Mon, 19 Dec 1994 09:52:27 -0500
Subject: "Blind" reviews are impossible
Message-ID: <199412191452.AA23787@PLANCK.CTAN.YALE.EDU>

Even a nonexpert reviewer can figure out who wrote a paper simply by
looking for citations of prior work.  The only way to guarantee a
"blind" review is to forbid authors from citing anything they've done
before, or insist on silly euphemisms when citing such publications.

--Ted


From jbower at smaug.bbb.caltech.edu  Mon Dec 19 15:44:10 1994
From: jbower at smaug.bbb.caltech.edu (jbower@smaug.bbb.caltech.edu)
Date: Mon, 19 Dec 94 12:44:10 PST
Subject: WCNN / INNS / or whatever
Message-ID: <9412192044.AA00733@smaug.bbb.caltech.edu>

In brief support of Jordon, I believe that it is well known that the
finances of both the WCNN and the INNS have been strange for years. It is
also not too surprising that one would confuse the two, same list of
participants, basically.  It is also well known that the "everyone who can
pay" approach taken by the organizers of most of the neural network
meetings has resulted in very poor signal to noise ratios.  The
historically more "old boy" approach of NIPS has meant a better quality
meeting, but less openness.  I wish Dave Touretzky luck in efforts to
change, there is little evidence that INNS or IEEE even recognize that
there is a problem.   

Two years ago, a considerable amount of pressure was placed on the
neurobiologists who have organized the computational neuroscience meetings
(CNS*92-94) to merge with the next INNS meeting (in this case).  The vote
by participants of CNS*93 not to, was essentially 100%. Other than the
comment that very little that goes on in any of these meetings has much to
do with neurobiology, the other reasons most often mentioned where the
above.


Jim Bower    

           ***************************************
                       James M. Bower
                     Division of Biology
                     Mail code:  216-76
                           Caltech
                     Pasadena, CA 91125
                      (818) 395-6817
                      (818) 449-0679 FAX

      NCSA Mosaic laboratory address:  http://www.bbb.caltech.edu/bowerlab
      NCSA Mosaic address for GENESIS:  http://www.bbb.caltech.edu/GENESIS




From maass at igi.tu-graz.ac.at  Mon Dec 19 15:48:06 1994
From: maass at igi.tu-graz.ac.at (Wolfgang Maass)
Date: Mon, 19 Dec 94 21:48:06 +0100
Subject: elitism at NIPS 
Message-ID: <199412192048.AA16441@figids01>


I have the impression that for theory-papers it would NOT be 
very benefitial if NIPS would change to "blind reviewing". 


From dtam at morticia.cnns.unt.edu  Mon Dec 19 16:28:15 1994
From: dtam at morticia.cnns.unt.edu (David Tam)
Date: Mon, 19 Dec 94 15:28:15 CST
Subject: elitism at NIPS
Message-ID: <199412192128.AA02352@morticia.cnns.unt.edu>

I think a totally honest system has to be doubly-open, i.e., the reviewers
names have to be attached to the referee's comments.  That will hold them
accountable for what they say.  That will keep them HONEST.  If they want to
give negative critisim, let it be known who they are.

As is now, it is not an open system -- It is blind one way, and open the
other way, and that makes the system so unfair, bias, and dictatorial.

As Scott Fahlman said, it is practically impossible to make it blind,
because we all know who does what line of work.  So, if it is not blind,
make it totally open, and have a repeal process, such that if the
referee is WRONG, there is a recourse!!

That's what a democratic process is all about -- to keep the system
honest by having an OPEN process with REPEAL recourse.  A closed system
is a sure way to breed corruption and dictatorship.

This response applies to the whole scientific review process in reviewing
papers and grants in general as well as in conferences.

David C. Tam
Center for Network Neuroscience
Dept. of Biological Sciences
University of North Texas



From shams at maxwell.hrl.hac.com  Mon Dec 19 20:39:28 1994
From: shams at maxwell.hrl.hac.com (Soheil Shams)
Date: Mon, 19 Dec 1994 17:39:28 -0800
Subject: Position Announcement
Message-ID: <9412200146.AA07164@maelstrom>

SIGNAL / IMAGE PROCESSING RESEARCH OPPORTUNITIES

Hughes Research Laboratories has an immediate opening for a Research Staff
Member to join a team of scientists in the Computational Intelligence and
Advanced Signal Processing Algorithms Project.  Team members in this
project have developed novel, state-of-the-art neural networks,
time-frequency transforms, and image compression algorithms for use in both
commercial and military applications.

The successful candidate will investigate advanced signal and image
processing techniques for multimedia compression, data fusion, and pattern
recognition applications.  Current work is focused on the application of
wavelets, neural networks, and computer vision techniques for data
compression and pattern recognition applications.  Specific duties will
include theoretical analysis, algorithm design, and software simulation.

Candidates are expected to have a Ph.D. in Electrical Engineering, Applied
Mathematics, or Computer Science.  Strong analytical skills and
demonstrated ability to perform creative research, along with experience in
signal and image processing, image or video compression, or neural networks
are required.  Practical experience with C or C++ is essential.   Good
communications and teamwork skills are keys to success.

Overlooking the Pacific Ocean and the coastal community of Malibu, the
Research Laboratories provides an ideal environment for you to make the
most of your scientific abilities.  Our organization offers a competitive
salary and benefits package.  

Additional information may be obtained from Lynn Ross.  For immediate
consideration, send your resume to:

Lynn W. Ross
Department RM
Hughes Research Laboratories
3011 Malibu Canyon Road
Malibu,  CA   90265
FAX: (310) 317-5651
Internet:  lross at msmail4.hac.com

Proof of legal right to work in the United States required.  An Equal
Opportunity Employer.



From wermter at nats2.informatik.uni-hamburg.de  Tue Dec 20 06:28:41 1994
From: wermter at nats2.informatik.uni-hamburg.de (Stefan Wermter)
Date: Tue, 20 Dec 94 12:28:41 +0100
Subject: IJCAI95-workshop: Learning for Natural Language Processing
Message-ID: <9412201128.AA01573@nats2.informatik.uni-hamburg.de>


---------------------------------------------------------------------------






                       CALL FOR PAPERS AND PARTICIPATION


                            IJCAI-95 Workshop on


         New Approaches to Learning for Natural Language Processing


       International Joint Conference on Artificial Intelligence (IJCAI-95)
                      Palais de Congres,  Montreal, Canada

                     currently scheduled for August 21, 1995



ORGANIZING COMMITTEE
--------------------

Stefan Wermter          Gabriele Scheler              Ellen Riloff
University of Hamburg   Technical University Munich   University of Utah


PROGRAM COMMITTEE
-----------------

Jaime Carbonell, Carnegie Mellon University, USA
Joachim Diederich, Queensland University of Technology, Australia
Georg Dorffner, University of Vienna, Austria
Jerry Feldman, ICSI, Berkeley, USA
Walther von Hahn, University of Hamburg, Germany
Aravind Joshi, University of Pennsylvania, USA
Ellen Riloff, University of Utah, USA
Gabriele Scheler, Technical University Munich, Germany
Stefan Wermter, University of Hamburg, Germany


WORKSHOP DESCRIPTION
--------------------

In the last few years, there has been a great deal of interest and activity  in
developing  new approaches to learning for natural language processing. Various
learning methods have been used, including

- connectionist methods/neural networks
- machine learning algorithms
- hybrid symbolic and subsymbolic methods
- statistical techniques
- corpus-based approaches.

In general, learning methods are designed to support  automated  knowledge  ac-
quisition,  fault  tolerance,  plausible  induction, and rule inferences. Using
learning methods for natural language processing is  especially  important  be-
cause  language learning is an enabling technology for many other language pro-
cessing problems, including  noisy speech/language integration, machine  trans-
lation,  and information retrieval. Different methods support language learning
to various degrees but, in general, learning is  important  for  building  more
flexible, scalable, adaptable, and portable natural language systems.

This workshop is of interest particularly at this time because systems built by
learning  methods  have reached a level where they can be applied to real-world
problems in natural language processing and where they  can  be  compared  with
more traditional encoding methods. The workshop will bring together researchers
from the US/Canada, Europe, Japan, Australia and other countries working on new
approaches to language learning.

The workshop will provide a forum for discussing  various  learning  approaches
for  supporting  natural  language processsing. In particular the workshop will
focus on questions like:

- How can we apply suitable existing learning methods for language processing?

- What new learning methods are needed for language processing and why?

- What language knowledge should be learned and why?

- What are similarities and differences between different approaches for
  language learning? (e.g., machine learning algorithms vs neural networks)

- What are strengths and limitations of learning rather than manual encoding?

- How can learning and encoding be combined in symbolic/connectionist systems?

- Which aspects of system architectures and knowledge engineering have to
  be considered? (e.g., modular, integrated, hybrid systems)

- What are successful applications of learning methods in various fields?
  (speech/language integration, machine translation, information retrieval)

- How can we evaluate learning methods using real-world language?
  (text, speech, dialogs, etc.)


WORKSHOP FORMAT
---------------

The workshop will provide a forum for the interactive  exchange  of  ideas  and
knowledge. Approximately 30-40 participants are expected and there will be time
for up to 15 presentations depending on the number and quality of paper contri-
butions received. Normal presentation length will be 15+5 minutes, leaving time
for direct questions after each talk. There may be a few invited talks of  25+5
minutes length.  In addition to prepared talks, there will be time for moderat-
ed discussions after two related sessions. Furthermore, the  moderated  discus-
sions  will provide an opportunity for an open exchange of comments, questions,
reactions, and opinions.

PUBLICATION
-----------

Workshop proceedings will be published by AAAI.  If  there  is  sufficient  in-
terest  of  the participants of the workshop there may be a possibility to pub-
lish the results of the workshop as a book.

REGISTRATION
------------

This workshop will take place directly before the general IJCAI-conference.  It
is  an IJCAI policy, that workshop participation is not possible without regis-
tration for the general conference.

SUBMISSIONS
-----------

All submissions will be refereed by the program committee and other experts  in
the  field.  Please submit 4 hardcopies AND a postscript file. The paper format
is the IJCAI95 format:  12pt article style latex, no more  than  43  lines,  15
pages  maximum,  including title, address and email address, abstract, figures,
references. Papers should fit to 8 1/2" x 11" size. Notifications will be  sent
by email to the first author.

Postscript files can be uploaded with anonymous ftp:

ftp nats4.informatik.uni-hamburg.de (134.100.10.104)
login: anonymous
password: <your email address>
cd incoming/ijcai95-workshop
binary
put <yourfile.Z or yourfile.gz>
quit

Hardcopies AND postscript files  must arrive not later than 24th February  1995
at the address below.

##############Submission Deadline:  24th February 1995
##############Notification Date: 24th March 1995
##############Camera ready Copy: 13th April 1995



Please send correspondence and submissions to:


################################################
Dr. Stefan Wermter
Department of Computer Science
University of Hamburg
Vogt-Koelln-Strasse 30
D-22527 Hamburg
Germany

phone:  +49 40 54715-531
fax:    +49 40 54715-515
e-mail: wermter at informatik.uni-hamburg.de
################################################


































From N.Sharkey at dcs.shef.ac.uk  Tue Dec 20 08:04:01 1994
From: N.Sharkey at dcs.shef.ac.uk (N.Sharkey@dcs.shef.ac.uk)
Date: Tue, 20 Dec 94 13:04:01 GMT
Subject: catastrophic interference of BP
In-Reply-To: Dr Neil Burgess - Anatomy UCL London's message of Mon, 19 Dec 94 14:08:28 +0000 <187786.9412191408@link-1.ts.bcc.ac.uk>
Message-ID: <9412201304.AA18623@entropy.dcs.shef.ac.uk>


>Studies of catastrophic interference in BP networks are
>interesting when considering such a network as a model of some human
>(or animal) memory system. 
>Is there any reason for doing that?
>Neil

BP has been used extensively in human cognitive and memory modelling and
so it is useful to know if "cashes in" in terms of recognition memory as
well.

The interference issue is also problematic for training online control processes
(e.g. in robotic coordination) where new training data is not necessarily accompanied
by all of the "old" data samples. If there is not a nice regular general function to
be extracted, then forgetting could be very severe. Problems also arise for BP in any
such online task that requires discrimination between what was in the training sets and
what was not.
noel



From gbugmann at school-of-computing.plymouth.ac.uk  Tue Dec 20 12:15:05 1994
From: gbugmann at school-of-computing.plymouth.ac.uk (Guido.Bugmann xtn 2566)
Date: Tue, 20 Dec 94 17:15:05 GMT
Subject: PhD Position available
Message-ID: <210.9412201715@subnode.soc.plym.ac.uk>


A PhD position is available immediately to work on the project "The
Plymouth Hand" at the University of Plymouth, UK.  This project comprises
an application of neural networks to control.

The aim of that project is to develop a robot/prosthetic hand with slippage
detection as a source of feedback for the control of the grip pressure.

A working prototype has been built which can lift a cylinder of solid steel
and also a hollow rolled sheet of A4 paper without deforming it.

Further work comprises:

1. Derivation of a practical specification for two systems using the
slippage detection adaptive grasping force technique, i.e. a robot gripper
and a prosthetic hand.

2. Kinematically model a range of proposed designs using virtual reality
software.

3. Validate appropriate kinematic models

4. Design, build and test an improved version of the existing device to
meet the specifications of a robot gripper. In particular:

a) Software environment will be investigated e.g.  assembly code, C, etc
and an appropriate one chosen and developed. New techniques such as rule
based (fuzzy logic) and neural networks will be investigated.

b) Design and development of the power electronics and drive system.

c) Specify and test the necessary microcontroller.

5) Same as 4) but as an adaptation of an existing NHS prostethic hand.
----------------------------------------------------

The salary in the standard range 5-6 Kpounds/year
for PhD students.

For further information, please
contact:

Paul Robinson
Dept. of Electrical and Electronic Engineering
University of Plymouth
Plymouth PL4 8AA
United Kingdom

Phone (+44) 1752 23 25 95 / 72
Fax (+44) 1752 23 25 83
-----------------------------------------------------


From ucganlb at ucl.ac.uk  Tue Dec 20 12:54:01 1994
From: ucganlb at ucl.ac.uk (Dr Neil Burgess - Anatomy UCL London)
Date: Tue, 20 Dec 94 17:54:01 +0000
Subject: catastrophic interference of BP 
Message-ID: <154923.9412201754@link-1.ts.bcc.ac.uk>

>> Studies of catastrophic interference in BP networks are
>> interesting when considering such a network as a model of some human
>> (or animal) memory system. 
>> Is there any reason for doing that?
>> Neil

> So far Jay McClelland has replied: his recently advertised 
> tech. report provides an example of useful consideration of cat. int. with repect to
> the possible existence of complimentary learning systems in the brain, and attempts 
> to distinguish between the specific properties of BP and the more general question.
> Japp Murre also replied that he partially addresses the problem in [1].

    Generally, I think that caution should be expressed in generalising
between artificial learning mechanisms. It may be that learning within a
system with fixed parameters, mediated by iterative minimisation of some
global attribute (e.g. sum-squared error) will tend to show interference
with `catastrophic' characteristics, although how it manifests itself will
depend on the details of the algorithm.
    But I would be surprised if other algorithms, e.g.  learning by
piecemeal constructive algorithms (where extra units are added for a
specific local task, such as [2]) behaved like that - so e.g. we might not
expect learning like song-learning in birds (associated with neural growth)
to necessarily show the same type of interference.
    I also suspect that the characteristics of interference differ between
iterative algorithms in which the `training set' must be presented many
times (e.g. BP) and `one-shot' learning algorithms (e.g. Hopfield model).
In the former case there is obviously a problem of how to interleave new
data into the training set, in the latter case there is no such problem,
and slight variations of the `Hebbian' learning rule can produce
imprinting, primacy, recency or combinations of the above [3].
    Given the availiblity of alternatives, it is not clear that BP should
always be the canonical choice for modelling learing and memory. It is
certainly not the easiest to motivate biologically.

Merry Christmas,

Neil

[1] report:  ftp://ftp.mrc-apu.cam.ac.uk/pub/nn/murre/hyper1.ps
[2] M. R. Frean, `The Upstart algorithm: A method for
  constructing and training feedforward neural networks',
  {\it Neural Computation,} {\bf 2}, 198-209 (1990).
[3] `List Learning in Neural Networks', Neil Burgess, J. L. Shapiro
  and M. A. Moore, {\it Network} {\bf 2} 399-422, 1991.



From nburgess at lbs.lon.ac.uk  Tue Dec 20 11:49:43 1994
From: nburgess at lbs.lon.ac.uk (nburgess@lbs.lon.ac.uk)
Date: Tue, 20 Dec 94 16:49:43 GMT
Subject: NNCM 95 Announcement
Message-ID: <199412201649.LAA04285@jupiter.lbs.lon.ac.uk>


                      ANNOUNCEMENT AND CALL FOR PAPERS

                                  NNCM-95        
                                    

                      THIRD INTERNATIONAL CONFERENCE ON
                    NEURAL NETWORKS IN THE CAPITAL MARKETS


                    Thursday-Friday, October 12-13, 1995
                with tutorials on Wednesday, October 11, 1995.

                    The Langham Hilton, London, England.



        Neural networks are now emerging as a major modelling methodology 
        in financial engineering. Because of the overwhelming interest of 
        the NNCM workshops held in London in 1993 and Pasadena in 1994, 
        the third annual NNCM conference will be held October 12-13, 1995, 
        in London. NNCM'95 takes a critical look at state of the art neural 
        network applications in finance.

        This is a research meeting where original, high-quality contribu-
        tions to the field are presented and discussed. In addition, a
        day of introductory tutorials (Wednesday, October 11) will be
        included to familiarise audiences of different backgrounds with
        financial engineering, and the mathematical aspects of the field.


        Application areas include:

        +     Bond and stock valuation and trading
        +     Foreign exchange rate prediction and trading
        +     Commodity price forecasting
        +     Risk management
        +     Tactical asset allocation
        +     Portfolio management
        +     Option Pricing
        +     Trading strategies


        Technical areas include but not limited to:

        +     Generalised least squares
        +     Robust model estimation
        +     Univariate time series analysis
        +     Multivariate data analysis
        +     Classification and ranking
        +     Pattern recognition
        +     Model selection
        +     Hypothesis testing and confidence intervals

        Instructions for Authors

        Authors who wish to present a paper should mail a copy of their 
        extended abstract (4 pages, single-sided, single-spaced) typed on 
        A4 (8.5" by 11") paper to the secretariat no later than May 31, 
        1995. Submissions will be refereed by no less than four referees 
        and authors will be notified on acceptance by 30 July 1995. Sep-
        arate registration is required using the attached registration 
        form. Authors are encouraged to submit abstracts as soon as pos-
        sible.

        Registration

        To register, complete the registration form and mail to the  sec-
        retariat. Please note that places are limited and will be allocated
        on a "first-come first-served" basis.

        Secretariat:

        For further information, please contact the NNCM-95 secretariat:

                Ms Busola Oguntula, London Business School
                Sussex Place, Regent's Park, London NW1 4SA, UK
                e-mail:  boguntula at lbs.lon.ac.uk
                phone (+44) (0171) 262 50 50 
                fax (+44) (0171) 724 78 75

        Location:

        The main conference will be held at The Langham Hilton which is
        situated near Regent's Park and is a short walk from Baker Street
        Underground Station. Further directions including a map will be
        sent to all registries.

        Programme Commitee

        Dr A. Refenes, London Business School (Chairman)
        Dr Y. Abu-Mostafa, Caltech
        Dr A. Atiya, Cairo University
        Dr N. Biggs, London School of Economics
        Dr D. Bunn, London Business School
        Dr M. Jabri, University of Sydney
        Dr B. LeBaron, University of Wisconsin
        Dr A. Lo, MIT Sloan School
        Dr J. Moody, Oregon Graduate Institute
        Dr C. Pedreira, Catholic University PUC-Rio
        Dr M. Steiner, Universitaet Munster
        Dr A. Timermann, University of California, San Diego
        Dr A. Weigend, University of Colorado
        Dr H. White, University of California, San Diego



        Hotel Accommodation:

        Convenient hotels include:

                The Langham Hilton
                1 Portland Place
                London W1N 4JA
                Tel: (+44) (0171) 636 10 00 
                Fax: (+44) (0171) 323 23 40

                Sherlock Holmes Hotel
                108 Baker Street, London NW1 1LB
                Tel: (+44) (0171) 486 61 61 
                Fax: (+44) (0171) 486 08 84

                The White House Hotel
                Albany St., Regent's Park, London NW1
                Tel: (+44) (0171) 387 12 00 
                Fax: (+44) (0171) 388 00 91


--------------------------Registration Form ----------------------------

                        NNCM-95 Registration Form

                    Third International Conference on
                  Neural Networks in the Capital Markets
                          October 12-13 1995

        Name:____________________________________________________

        Affiliation:_____________________________________________

        Mailing Address: ________________________________________

        _________________________________________________________

        Telephone:_______________________________________________

  ****Please circle the applicable fees and write the total below****


        Main Conference (October 12-13):
                                                         (British Pounds)
                Registration fee                                450                          
                
                Discounted fee for academicians                 250
                (letter on university letterhead required)      

                Discounted fee for full-time students           100
                (letter from registrar or faculty advisor
                required)                                               

        Tutorials (October 11):

        You must be registered for the main conference in order to 
        register for the tutorials.

                                                         (British Pounds)
                Morning Session Only                            100

                Afternoon Session Only                          100

                Both Sessions                                   150

                Full-time students                               50
                (letter from registrar or faculty advisor 
                required)

        TOTAL:  _________                                

        Payment may be made by: (please tick)
        ____ Check payable to London Business School
        ____ VISA         ____Access         ____American Express

        Card Number:___________________________________




--


From pah at unixg.ubc.ca  Tue Dec 20 14:39:18 1994
From: pah at unixg.ubc.ca (Phil A. Hetherington)
Date: Tue, 20 Dec 1994 11:39:18 -0800 (PST)
Subject: catastrophic interference of BP
In-Reply-To: <187786.9412191408@link-1.ts.bcc.ac.uk>
Message-ID: <Pine.SUN.3.91.941220105036.16867B-100000@unixg.ubc.ca>


Neil Burgess wrote:

> Studies of catastrophic interference in BP networks are
> interesting when considering such a network as a model of some human
> (or animal) memory system. 
> Is there any reason for doing that?

Of course.  Network models have properties such as distributed
representations, generalization, interference, content addressability,
etc., that are also true of animal and human memory.  They provide an
alternative framework for construing memory processes that is superior to
box and arrow modeling primarily because they can be 'lesioned' as in
animal experiments or as found in human patients and because they are
'executable'.  Because they are executable (i.e., perform a function) the
effects of these lesions and other parameter manipulations on the
performance of the network can be observed.  There are now many
alternative supervised learning algorithms available, but there are still 
many reasons to continue to study this one. It is most certainly
true that both humans and animals learn via feedback provided by the
effect of erroneous behavior--a process analagous to back prop. 
Unsupervised learning algorithms such as competetive learning do not give
you this.  Unsupervised learning algorithms are not easily executable--you
can't get them to do many simple 'behaviors', like plot trajectories from
starting locations to multiple goals, as in Neil's models.  Of course, any
supervised learning algorithm will confer the ability to train the net to
perform, but there are a couple, mostly pragmatic, reasons to stick with
back prop for now.  Primarily because of the availablity of the
McClelland & Rumelhart books and program disks, back prop is most easily
available and most commonly used.  Back prop already provides the engine
in countless models already published.  Gaining an understanding of the
behavior of this algorithm will enable a better understanding of the flaws
of these already published models.  Thus, its not so much that you *would
want to* use the algorithm, it is that it has already been used.  Lets
understand why we are discarding it before we do so. 

Phil Hetherington
pah at unixg.ubc.ca


From minton at ptolemy-ethernet.arc.nasa.gov  Tue Dec 20 17:39:00 1994
From: minton at ptolemy-ethernet.arc.nasa.gov (Steve Minton)
Date: Tue, 20 Dec 94 14:39:00 PST
Subject: Learning Article
Message-ID: <9412202239.AA09677@ptolemy.arc.nasa.gov>

Readers of this group may be interested in the following article, which
was just published in the Journal of Artificial Intelligence Research
(a journal which is available both online and in print).

Buntine, W.L. (1994)
  "Operations for Learning with Graphical Models", Volume 2, pages 159-225
   Postscript: volume2/buntine94a.ps (1.53M)
               compressed, volume2/buntine94a.ps.Z (568K)

  Abstract: This paper is a multidisciplinary review of empirical,
  statistical learning from a graphical model perspective.  Well-known
  examples of graphical models include Bayesian networks, directed
  graphs representing a Markov chain, and undirected networks
  representing a Markov field.  These graphical models are extended to
  model data analysis and empirical learning using the notation of
  plates.  Graphical operations for simplifying and manipulating a
  problem are provided including decomposition, differentiation, and the
  manipulation of probability models from the exponential family.  Two
  standard algorithm schemas for learning are reviewed in a graphical
  framework: Gibbs sampling and the expectation maximization algorithm.
  Using these operations and schemas, some popular algorithms can be
  synthesized from their graphical specification.  This includes
  versions of linear regression, techniques for feed-forward networks,
  and learning Gaussian and discrete Bayesian networks from data.  The
  paper concludes by sketching some implications for data analysis and
  summarizing how some popular algorithms fall within the framework
  presented.
  
  The main original contributions here are the decomposition techniques
  and the demonstration that graphical models provide a framework for
  understanding and developing complex learning algorithms.


The PostScript file is available via:
   
 -- comp.ai.jair.papers

 -- World Wide Web: The URL for our World Wide Web server is
       http://www.cs.washington.edu/research/jair/home.html

 -- Anonymous FTP from either of the two sites below:
      CMU:   p.gp.cs.cmu.edu        directory: /usr/jair/pub/volume2
      Genoa: ftp.mrg.dist.unige.it  directory:  pub/jair/pub/volume2

 -- automated email. Send mail to jair at cs.cmu.edu or jair at ftp.mrg.dist.unige.it
    with the subject AUTORESPOND, and the body GET VOLUME2/BUNTINE94A.PS
    (either upper or lowercase is fine). 
    Note: Your mailer might find this file too large to handle.
          (The compressed version of this paper cannot be mailed.)

 -- JAIR Gopher server: At p.gp.cs.cmu.edu, port 70. 

For more information about JAIR, check out our WWW or FTP sites, or
send electronic mail to jair at cs.cmu.edu with the subject AUTORESPOND
and the message body HELP, or contact jair-ed at ptolemy.arc.nasa.gov.




From gary at cs.ucsd.edu  Tue Dec 20 18:29:18 1994
From: gary at cs.ucsd.edu (Gary Cottrell)
Date: Tue, 20 Dec 94 15:29:18 -0800
Subject: "Blind" reviews are impossible
Message-ID: <9412202329.AA18047@desi>


Right, silly euphemisms work, or simply saying, "prior work on this
by cottrell, etc." The year I was on the NIPS PC, I was also on
the PC for ACL and AAAI. ACL used blind reviewing. I was sure I was
reviewing a paper by Ken Church, and I was wrong. There was only one
paper of the 15 or so I reviewed that I had any idea who it was.

gary cottrell


From pf2 at st-andrews.ac.uk  Tue Dec 20 17:07:07 1994
From: pf2 at st-andrews.ac.uk (Peter Foldiak)
Date: Tue, 20 Dec 94 22:07:07 GMT
Subject: research assistant job - computational neuroscience
Message-ID: <8312.9412202207@psych.st-andrews.ac.uk>

     University of St Andrews, School of Psychology
     RESEARCH ASSISTANT
     Computational Neuroscience

A one-year research assistantship is available for research
on a novel application of computational methods to the
neurophysiology of the primate visual system with
Dr Peter Foldiak, in collaboration with Dr David Perrett.

The assistant will be involved in the development of software and
experimental procedures. Training in computer programming (C, UNIX),
and preferably in some of the following areas is preferred:
mathematics (optimisation), neural networks, genetic algorithms,
computer graphics (on SGI), vision- or neuroscience.

Salary will be at the appropriate point on the 1B scale for
Research Staff (GBP 13941-17813).
Application forms and further particulars are available from
Personnel Services, University of St Andrews, KY16 9AJ, U.K.
tel: +44 334 462567, (out of hours +44 334 462571) or by fax:
+44 334 462570, to whom completed forms accompanied by a letter
of application should be returned to arrive not later than
Monday, 30 January 1995. Please quote reference number: SH/APS0838.

The University operates an Equal Opportunities Policy.


From jaap.murre at mrc-apu.cam.ac.uk  Wed Dec 21 12:50:03 1994
From: jaap.murre at mrc-apu.cam.ac.uk (Jaap Murre)
Date: Wed, 21 Dec 94 17:50:03 GMT
Subject: sequential inteference
Message-ID: <9412211750.AA26645@rigel.mrc-apu.cam.ac.uk>


In response to recent comments by Neil Burgess, Bob French, Phil
Hetherington, Noel Sharkey, Jay McClelland and others on 'catastrophic
interference', I think it is important to establish that 'vanilla
backpropagation' has by now been eliminated as a valid model of human
learning and memory, its implausible learning transfer being the most
obvious failure. An important reason for this failure can be found in the
nature of the hidden layer representations.
      Catastrophic interference and hypertransfer (i.e., excessive positive
transfer, see Murre, in press a) are both sides of the same coin: hidden-
layer representations in backpropagation bear little relationship to
orthogonalities of input patterns. In the case of interference experiments in
humans the following result is well established: If the input patterns
(stimuli) in two consecutive learning sets A and B are different than there
will be neither interference nor positive transfer (e.g., Osgood, 1949). This
is not the case in backpropagation: orthogonality of the stimuli has no
effect on the reduction of interference. The reason for this is that the
hidden-layer representations are always about equally overlapping *no
matter what the input stimuli are*. Interference and transfer in
backpropagation are only psychologically implausible with respect to its
indifference towards the structure of the input stimuli. Surprisingly enough,
two-layer networks (i.e., with a normal delta-rule, with one layer of
weights) do not suffer from this problem and are in fact well able to model
human interference (see Murre, in press a).
      Having said this, it is also important to consider the various variant
models of error-correcting learning that have been developed recently and
that use backpropagation as a starting point. These models have very
plausible characteristics with respect to learning and categorization in
humans: Kruschke (1990), Gluck (1991), Gluck and Bower (1988, 1990),
Nosofsky, Kruschke, and McKinley (1992), Shanks and Gluck (1994). In
addition, there are now many variant models of backpropagation that do
not suffer from catastrophic interference. These have already been
mentioned in this discussion, so I will not repeat them here.
      From a biological point of view backpropagation is certainly
implausible but it has been show useful in inferring biological plausible
parameters (e.g., Lockery et al., 1989; Zipser and Anderson, 1988).

I can think of several reasons why backpropagation continues to attract
researchers:

1. It is easy to understand.

2. Many simulators are available (see Murre, in press b).

3. It has been shown to approximate all 'well-behaved' functions (e.g.,
Hornik, Stinchcombe, and White, 1989) and thus is often felt to qualify as
a generic learning mechanism. In particular, it can learn non-linearly
separable pattern sets.

4. It has only a few parameters and these are not very critical for the final
results.

5. It possesses most of the basic elements of a 'prototypical' neural
network: distributed representations, graceful degradation, pattern
completion, and adequate generalization of learned behavior. 

I do not myself think that these reasons form necessarily a sufficient
motivation for using backpropagation, and I indeed prefer to work on other
types of learning methods and neural networks. Perhaps, by investigating
the limitations of backpropagation, necessary minimal improvements may
become clear, so that we can replace it - in its leading role - by either a
more plausible variant algorithm, or by a completely different learning
method. 

Merry Christmas,

                      -- Jaap Murre



References

Gluck, M.A. (1991). Stimulus generalization and representation in adaptive
      network models of category learning. Psychological Science, 2, 50-
      55.
Gluck, M.A., & G.H. Bower (1988). From conditioning to category
      learning: an adaptive network model. Journal of Experimental
      Psychology: General, 117, 227-247.
Gluck, M.A., & G.H. Bower (1990). Component and pattern information in
      adaptive networks, Journal of Experimental Psychology: General,
      119, 105-109.
Kruschke, J.K. (1990). ALCOVE: an exemplar-based connectionist model
      of category learning. Psychological Review, 99, 22-44.
Lockery, S.R., G. Wittenberg, W.B. Kristan, Jr., & Garrison W. Cottrell
      (1989). Function of identified interneurons in the leech elucidated
      using neural networks trained by back-propagation. Nature, 340, 468-
      471.
Murre, J.M.J. (in press a). Transfer of learning in backpropagation and in
      related neural network models. In: J. Levy, D. Bairaktaris, J.
      Bullinaria, & P. Cairns (Eds.), Connectionist Models of Memory and
      Language. London: UCL Press. (In our ftp site:
      ftp://ftp.mrc-apu.cam.ac.uk/pub/nn/murre/hyper1.ps) 
Murre, J.M.J., (in press b). Neurosimulators. In: M.A. Arbib (Ed.),
      Handbook of Brain Research and Neural Networks, Cambridge, MA:
      MIT Press. (In our ftp site:
      ftp://ftp.mrc-apu.cam.ac.uk/pub/nn/murre/neurosim1.ps).
Nosofsky, R.M., J.K. Kruschke, and S. C. McKinley (1992). Combining
      exemplar-based category representations and connectionist learning
      rules. Journal of Experimental Psychology: Learning, Memory, and
      Cognition, 18, 211-233.
Osgood, C.E. (1949). The similarity paradox in human learning: a
      resolution. Psychological Review, 56, 132-143.
Shanks, D.R., & M.A. Gluck (1994). Tests of an adaptive network model
      for the identification and categorization of continuous-dimension
      stimuli. Connection Science, 6, 59-89.
Zipser, D., & R.A. Anderson (1988). A back-propagation programmed
      network that simulates response properties of a subset of posterior
      parietal neurons. Nature, 331, 679-684.



From jras at uned.es  Wed Dec 21 13:12:00 1994
From: jras at uned.es (Jose Ramon Alvarez Sanchez)
Date: 21 Dec 94 19:12 +0100
Subject: CFP: W.S. McCulloch, 25 years in Memoriam
Message-ID: <201*jras@uned.es>


                      Preliminary Call For Papers

    CENTRO INTERNACIONAL DE INVESTIGACION EN CIENCIAS DE LA COMPUTACION
                UNIVERSIDAD DE LAS PALMAS DE GRAN CANARIA


                  W.S. McCulloch: 25 years in Memoriam

      INTERNATIONAL CONFERENCE ON BRAIN PROCESSES, THEORIES AND MODELS

             Las Palmas, Gran Canaria, Spain. Nov 12-17, 1995.


Conference Chairman: Moreno-Diaz, R. (Spain)
Org. Comm. Chairman: Mira-Mira, J. (Spain)


Scientific Advisory Comm. (preliminary):

Amari SI (Japan), Anderson J (USA), Arbib M (USA), Beer S (Canada),
Belmonte C (Spain), Blum M (USA), Braitenberg V (Germany), Cull P (USA),
DaFonseca JL (Portugal), Eckorn R (Germany), Harth E (USA),
Herault J (France), Jain LC (Australia), Kauffman S (USA), Kilmer W (USA),
Leibovic KN (USA), Lettvin J (USA), Malsburg C. von der (Germany),
Maturana H(Chile), McCleland J (USA), Mira-Mira J (Spain), Papert S (USA),
Pichler F (Austria), Ricciardi L (Italy), Sato S (Japan),  
Vitozz E (Switzerland), Von Foester H (USA).

Organizing Comm.:

Alvarez JR (Spain), Cabestany J (Spain), Delgado A (Spain), 
Krasner J (USA), Moreno-Diaz jr R (Spain), Pierce A (USA),
Prieto A (Spain), Sanchez JV (Spain), Suarez-Araujo C (Spain).

Organization Staff:

Alonso-Garcia MT (Spain), Perez Ruiz M (Spain).


                             CALL FOR PAPERS

     The conference will consist in a series of invited lectures by leading
scientists icluding M. Arbib, H. Maturana, J. Lettvin and S. Papert related to
the life and work of W.S.McCulloch, an open forum of paper sessions on the
topics listed below and a workshop on the "Embodiments of Mind at the end of
the Century".

Relevant topics include:

      Anatomical, Physiological, Biochemical and Biophysical levels.
      Natural and artificial neural networks.
      Mathematics, systems theory and global properties of the nervous system.
      Conceptual and formal tools in brain function modelling.
      Hybrid systems: symbolic-connectionist links.
      Plasticity, reliability, learning and memory.
      Implications of the work of WS McCulloch for philosophy and psychology.
      Reverse engineering and neurophysiology.

    
     All accepted papers will be published by the MIT Press in the 
pre-conferece proceedings. Those authors will be sent specific instructions
and guidelines. 

     Invited lectures and selected papers will be published by The MIT Press
in a post-conference book.

     Three copies of the intended papers should be sent, prepared according
to the following instructions for authors no later than May 30, 1995 to:

     Prof. Jose Mira-Mira
     Dpto. Informatica y Automatica-UNED
     Senda del Rey s/n
     28040 MADRID-SPAIN

     Voice: +34 (1) 398 7155
     Fax: +34 (1) 398 6697
     e-mail: <jose.mira at uned.es>

     Please include your fax, e-mail, telephone for quick contact
and notification of acceptance.


INSTRUCTIONS FOR AUTHORS

     Authors should submit three copies of full intended papers, not exceeding
eight pages of DIN A4 or 8.5 by 11 inch paper, including figures, tables and
references, in English.

     The centered heading must include:
	
         Title in capitals.
         Name(s) of author(s).
         Address(es) of author(s)
         A 10 lines abstract.

     Three blank lines should be left between each of the above items and four
between the heading and the body of the paper, 1.6 cm left, right, top and
bottom margins, single-spaced and not exceeding the 8-page limit.

     One additional DIN A-4 must be enclosed with the following organizational
data:

     Title and author(s) name(s).
     A list of five keywords.
     A reference to the Topics the paper relates to.
     Postal address, phone, fax and e-mail if available.

	
     All accepted papers will be published in the conference proceedings.


IMPORTANT DATES

      Final call and additonal info: March 1995
      Final date of submission: May 30, 1995
      Notification of acceptance: July 1, 1995
      Conference: November 12-17, 1995


REGISTRATION FORM

------------------------------------------------------------------------
Last Name:
First name
Organization/University:
Address:
Phone:
Fax:
E-mail:

Do you plan to submit a paper:
Tentative title:

Payment:
     Before April 30:
       - Bank draft for 50.000 pts payable on a Spanish Bank to
         FUNDACION UNIVERSITARIA DE LAS PALMAS: MCCULLOCH 95.
       - Money Transfer of 50.000 pts to BEX account No.
         0104 0336 75 0307007836
     After April 30:
       - Late registration fee of 60.000 pts at the conference desk.
------------------------------------------------------------------------
     Please send this registration form to:

     Mrs. Maria Teresa Alonso-Garcia
     CIICC-Universidad de Las Palmas
     Campus de Tafira-Edf. de Informatica
     35017 Las Palmas, SPAIN




From tetewsky at lima.psych.mcgill.ca  Wed Dec 21 15:30:55 1994
From: tetewsky at lima.psych.mcgill.ca (Sheldon Tetewsky)
Date: Wed, 21 Dec 1994 15:30:55 -0500
Subject: catastrophic interference
Message-ID: <199412212030.PAA03615@lima.psych.mcgill.ca>

Neil Burgess recently wrote that
>studies of catastrophic interference in BP networks are interesting when
>considering such a network as a model of some human (or animal) memory
>system.
However, he also questioned whether or not there was
>..any reason for doing that.

In the Tetewsky, Shultz, and Buckingham study ("Assessing interference and
savings in connectionist models of a sequential recognition memory task"),
referenced in Bob French's recent posting, we did some simulations and 
experiments indicating that neural networks can provide a good account of
human performance in a simple recognition memory task when memory is
assessed in terms of savings scores.  The paper is in progress and will
be announced soon. In the meantime, I'll just mention a few relevant
details.

Our work was motivated by Scott Fahlman's idea that his cascade-correlation
algorithm (CC) has certain inherent design features that should give it an
advantage over BP when it comes to dealing with the problem of catastrophic
interference.  We tested this idea by using an encoder version of CC to
model recognition memory.  Results indicated that in contrast to previous
findings, retroactive interference is not that serious a problem for BP
when memory is measured in terms of savings scores. However, CC also
produced a significant increase in savings, relative to BP.  Aside from
this difference in the magnitude of savings, we also have evidence that CC
was better than BP at accounting for the relative number of trials that
subjects used in the different phases of learning.

-- Sheldon Tetewsky 


From rsun at cs.ua.edu  Wed Dec 21 17:05:13 1994
From: rsun at cs.ua.edu (Ron Sun)
Date: Wed, 21 Dec 1994 16:05:13 -0600
Subject: No subject
Message-ID: <9412212205.AA29941@athos.cs.ua.edu>


=========================================================================
                 Call For Papers and Participation

                      The IJCAI Workshop on
                   Connectionist-Symbolic Integration: 
                      From Unified to Hybrid Approaches



                        to be held at IJCAI'95 
                          Montreal, Canada
                          August 19-20, 1995
-------------------------------------------------------------------------



There has been a considerable amount of research in integrating
connectionist and symbolic processing.  While such an approach has
clear advantages, it also encounters serious difficulties and
challenges.  Therefore, various models and ideas have been proposed to
address various problems and aspects in this integration. There is a
growing interest from many segments of the AI community, ranging from
expert systems, to cognitive modeling, to logical reasoning.

Two major trends can be identified in the state of the art: these are
the unified or purely and the hybrid approaches to integration.
Whereas the purely connectionist ("connectionist-to-the-top") approach
claims that complex symbol processing functionalities can be achieved
via neural networks alone, the hybrid approach is premised on the
complementarity of the two paradigms and aims at their synergistic
combination in systems comprising both neural and symbolic components.
In fact, these trends can be viewed as two ends of an entire spectrum.

Up till now, overall, there is still relatively little work in
comparing and combining these fairly isolated efforts. This workshop
will provide a forum for discussions and exchanges of ideas in this
area, to foster cooperative work.  The workshop will tackle important
issues in integrating connectionist and symbolic processing.


A tentative Schedule
---------------------

Day 1:

A. Introduction:

* Invited talks  
  These talks will provide an overview of the field and set the tone for 
  ensuing discussions. 
* Theoretical foundations for integrating connectionist and symbolic 
  processing

B. Definition of the two approaches: 

* Do they exhaust the space of current research in connectionist-symbolic 
  integration, or is there room for additional categories?
* How do we compare the unified and hybrid approaches?
* Do the unified and hybrid approaches constitute a clearcut dichotomy or
  are they just endpoints of a continuum?
* What class of processes and problems is well-suited to unified or hybrid 
  integration? The relevant motivations and  objectives.
* What type of model is suitable for what type of application? Enumerate 
  viable target domains.

C. State of the art:

* Recent or ongoing theoretical or experimental research work 
* Implemented models belonging to either the unified or hybrid approach
* Practical applications of both types of systems

Research addressing key issues concerning:

* the unified approach: theoretical or practical issues involving 
  systematicity, compositionality and variable binding, biologically 
  inspired models, connectionist knowledge representation, other 
  high-level connectionist models.

* the hybrid approach: modes and methods of coupling, task sharing between 
  various components of a hybrid system, knowledge representation and sharing.

* both: commonsense reasoning, natural language processing, analogical 
  reasoning, and more generally applications of unified and hybrid models.


Day 2:

D. Cognitive Aspects:

* Cognitive plausibility and relations to other AI paradigms
* In cognitive modeling, why should we integrate connectionist and symbolic 
  processing?
* Is there a clear cognitive rationale for such integration?  (we may
  need to examine in detail some typical areas, such as commonsense reasoning,
  and natural language processing)
* Is there psychological and/or biological evidence for existing models? 
  If so, what is it?

E. Open research issues:

* Can we now propose a common terminology with precise definitions for both 
  approaches to connectionist-symbolic integration and for the location on 
  the continuum?
* How far can unified systems go? Can unified models be supplemented by 
  hybrid models? Can hybrid models be supplanted by unified models?
* Limitations and barriers faced by both approaches
* What breakthroughs are needed for both approaches?
* Is it possible to synthesize various existing models? 


Workshop format
---------------
- panel discussions 
- mini-group discussions: participants will break into groups of 7/8
  to discuss a given theme; group leaders will then form a panel to
  report on group discussions and attempt a synthesis with audience
  participation 
- interactive talks: this is a novel type of oral presentation we will 
  experiment with. Instead of a classical presentation, the
  speaker will present a problem or issue and give a brief statement
  of his personal stand (5 min) to launch discussions which he will
  then moderate and conclude.
- classical slide talks followed by Q/A and discussions.


Workshop Co-chairs: 
-------------------
Frederic Alexandre, Crin-Cnrs/Inria-Lorraine
Ron Sun,   The University of Alabama 

Organizing Committee:
---------------------
John Barnden, New Mexico State University 
Steve Gallant,  Belmont Research Inc.       
Larry Medsker,  American University
Christian Pellegrini, University of Geneva
Noel Sharkey,  Sheffield University

Program Committee:
------------------
Lawrence Bookman (Sun Laboratory, USA)
Michael Dyer (UCLA, USA)
Wolfgang Ertel (FRW, Germany)
LiMin Fu (University of Florida, USA)
Jose Gonzalez-Cristobal (UPM, Spain)
Ruben Gonzalez-Rubio (University of Sherbrooke, Canada)
Jean-Paul Haton (Crin-Inria, France)
Melanie Hilario (University of Geneva, Switzerland)
Abderrahim Labbi (IMAG, France)
Ronald Yager (Iona College, USA)


Schedule:
---------
- The submission deadline for participants is February 1, 1995.
- The authors and potential participants will be notified the acceptance 
  decision by March 15, 1995.
- The camera-ready copies of working notes papers will be due on April 15, 1995


Submission:
-----------
- If you wish to present a talk, specify the preferred type of presentation 
  (classical or interactive talk) and submit 5 copies of an
  extended abstract (within the limit of 5-7 pages) to:

Prof. Ron Sun   
Department of Computer Science
The University of Alabama 
Tuscaloosa, AL 35487
(205) 348-6363

- If you only wish to attend the workshop, send 5 copies of a short
  (no more than one page) description of your interest to the same address 
  above.

- Please be sure to include your e-mail address in all submissions.



From marks at u.washington.edu  Thu Dec 22 00:55:40 1994
From: marks at u.washington.edu (Robert Marks)
Date: Wed, 21 Dec 94 21:55:40 -0800
Subject: Neural Networks in Russia
Message-ID: <9412220555.AA24404@carson.u.washington.edu>


For Connectionists:

              The 2-nd International Symposium
           on Neuroinformatics and Neurocomputers
                   Rostov-on-Don, RUSSIA
                   September 20-23, 1995

     Organized by Russian Neural Network Society (RNNS)

                          and

 A.B. Kogan Research Institute for Neurocybernetics (KRINC)

                   in co-operation with

     Institute of Electrical and Electronics Enginieers
             Neural Networks Council (IEEE NNC)


                   First Call for Papers

Research in Neuroinformatics and Neurocomputing continued in
Russia after  the research  was deflated  in the west in the
1970's.   The research sophistication in neural networks, as
a   result,  is   quite   advanced   in   Russia.  The first
international  RNNS/IEEE  Symposium, held  in October  1992,
proved  to  be  a  highly  successful  forum  for  a diverse
international  interchange  of   fresh  and  novel  research
results.    The    second   International    Symposium    on
Neuroinformatics   and   Neurocomputers   is   built on this
remarkable success.

The symposium  focus is  on the  neuroscience,  mathematics,
physics,  engineering  and  design  of  neuroinformatic  and
neurocomputing systems.

Rostov-on-Don, the  location of the Symposium, is about 1000
km south  of Moscow  on the  scenic Don  river.   The Don is
commonly identified  as the  boundary between the continents
of Europe  and Asia.   Rostov  is the home of the A.B. Kogan
Research Institute  for  Neurocybernetics  at  Rostov  State
University -  one of  the premier  neural  network  research
centers in Russia.

Papers for  the Symposium  should be  sent  in  CAMERA-READY
FORM, NOT  EXEEDING 8  PAGES in A4  format  to   the Program
Committee Co-Chair Alexander A. Frolov.   Two copies  of the
paper should  be submitted.

The deadline for submission is 15 MARCH, 1995.

Notification of acceptance will be sent on or before 15 May,
1995.




                    SYMPOSIUM COMMITTEE

GENERAL CHAIR

     Witali L. Dunin-Barkowski, Dr. Sci.,
     The 2-nd International Symposium
     on Neuroinformatics and Neurocomputers,
     Symposium Chair
     President of the Russian Neural Network Society,
     A.B. Kogan Research Institute for Neurocybernetics
     Rostov State University
     194/1 Stachka avenue,
     344104, Rostov-on-Don, Russia
     Tel: +7-8632-28-0588,
     Fax: +7-8632-28-0367
     E-mail: wldb at krinc.rostov-na-donu.su


PROGRAM COMMITTEE CO-CHAIRS

     Professor Alexander A. Frolov,
     The 2-nd International Symposium on Neuroinformatics and
        Neurocomputers,
     Program Co-Chair
     5a Butlerov str.
     Higher Nervous Activity and Neurophysiology Institute
     Russian Academy of Science
     117220, Moscow, RUSSIA.

     Professor Robert J. Marks II
     Program Co-Chair
     The 2-nd International Symposium on Neuroinformatics and
        Neurocomputers
     University of Washington
     Department of Electrical Engineering
     c/o 1131 199th Street S.W., Suite N
     Lynnwood, WA 98036-7138
     WA, USA.

Other information is available from the Symposium Committee.





From david at cns.ed.ac.uk  Thu Dec 22 03:58:07 1994
From: david at cns.ed.ac.uk (David Willshaw)
Date: Thu, 22 Dec 1994 08:58:07 GMT
Subject: elitism at NIPS
Message-ID: <199412220858.IAA09676@dumbo.cns.ed.ac.uk>


1. Even though the practice of removing the authors' names from
submitted papers before sending to referees is not perfect, in my view
it certainly helps to make the refereeing process less biased.

2. I would also support the suggestion that the cohort of referees be
broadened. One way of doing this is to make it more international. In
the list of 174 referees given in this year's programme of Abstracts,
I estimated that 84% (146) were from the USA and Canada. Only 6
referees were from Germany, 4 from France and 3 from the UK, for
example.

David



From trevor at media-lab.media.mit.edu  Thu Dec 22 11:01:55 1994
From: trevor at media-lab.media.mit.edu (Trevor Darrell)
Date: Thu, 22 Dec 94 11:01:55 EST
Subject: "Blind" reviews are impossible
In-Reply-To: ted@spencer.ctan.yale.edu's message of Mon, 19 Dec 1994 09:52:27 -0500 <199412191452.AA23787@PLANCK.CTAN.YALE.EDU>
Message-ID: <9412221601.AA06532@marblearch.media.mit.edu>

   Delivery-Date: Thu, 22 Dec 94 10:24:24 -0500
   Date: Mon, 19 Dec 1994 09:52:27 -0500
   From: ted at spencer.ctan.yale.edu

   Even a nonexpert reviewer can figure out who wrote a paper simply by
   looking for citations of prior work.  The only way to guarantee a
   "blind" review is to forbid authors from citing anything they've done
   before, or insist on silly euphemisms when citing such publications.

In computer vision papers have been reviewed blind at the major
conferences for the past several years. Authors are encouraged to
reference themselves in the third person. There seem to be
few complaints about the system. 

While it is sometimes possible to discern that a paper comes from a
particular school or group, it is usually impossible to know who was
the first author. And one can never be sure that the paper is not from
some other (unknown) person who has written the paper building
directly on the tradition of another group, and thus uses their
terminology but does not actually work there. In any case, can there
really be a disadvantage to blind reviewing? (It seems a weak point
indeed to say the correctness of papers in NIPS is assurred by their
authorship!)

Papers are not blind at the program committee level, so any agregious
mistakes brought on by blind reviewing (?) can be still corrected there. 

Just my $0.02 from the CV perspective...
--trevor


From trevor at mallet.Stanford.EDU  Thu Dec 22 18:10:03 1994
From: trevor at mallet.Stanford.EDU (Trevor Hastie)
Date: Thu, 22 Dec 1994 15:10:03 -0800
Subject: Report Announcement
Message-ID: <199412222310.PAA09264@mallet.Stanford.EDU>

The following report is available via anonymous ftp or Mosaic:

ftp://playfair.stanford.edu/pub/reports/hastie/dann.ps.Z

	Discriminant Adaptive Nearest Neighbor Classification

		Trevor Hastie and Robert Tibshirani

We propose an adaptive nearest neighbor rule that uses local
discriminant information to estimate an effective metric for
classification.  We also propose a method for global dimension
reduction, that combines local dimension information.  In a number of
examples, the methods demonstrate the potential for substantial
improvements over nearest neighbor classification.



From mozer at neuron.cs.colorado.edu  Thu Dec 22 18:32:47 1994
From: mozer at neuron.cs.colorado.edu (Michael C. Mozer)
Date: Thu, 22 Dec 1994 16:32:47 -0700
Subject: NIPS, blind reviewing, and elitism
Message-ID: <199412222332.QAA08554@neuron.cs.colorado.edu>

As NIPS*95 program chair, I want to respond to the issue of blind reviewing.
We have considered this idea over the past few months, and in balance the 
costs seem to outweigh the benefits.  Most of the arguments for and against 
blind reviewing were stated clearly in earlier messages.

NIPS is an elite conference in that researchers tend to self-select their best
work for submission, and even then only 25-30% of the submissions are accepted.
Further, researchers who do good work one year and have papers accepted are 
likely to do good work in the future, so it is not surprising that there is 
a core group of consistent contributors to the conference, even without any 
bias.  (Serving on the program committee before, I witnessed an opposite 
bias -- a bias against accepting multiple papers by an individual who 
had several strong submissions, and against awarding a talk to the same 
individual in successive years.)

Prior to 1994, there was likely some validity to the perception that NIPS 
"insiders" had an edge.  As a reviewer, it was difficult to evaluate work
solely on the basis of extended abstracts; in borderline cases, it helped to 
know the authors' track records.  After Dave Touretzky changed the submission
format to full papers in 1994, reviewers and program committee members to
whom I've spoken seemed satisfied that the reviewing process was objective.  
I suspect that the perception of unfairness may linger for a few years, but 
any such reality was squelched by the new submission format.

If you felt the reviewing process was not objective in 1994, I would like to 
hear your story.  (It is possible to communicate anonymously; to find out more 
about this service, send mail to help at anon.penet.fi.)  Several points of note 
before expressing a complaint:  (1) Many good submissions were ultimately 
rejected, simply because the submission pool tends to be of high quality and 
the number of accepted papers is limited by a maximum page count of the 
proceedings volume.  (2) Most of the people who have served as program and 
general chairs at NIPS have had papers rejected, including the current chairs!
The reviewers are the primary decision makers.  (3) Reviewers are not always
as competent and dilligent as one might like, although NIPS uses three 
reviewers per paper--plus an area chair to arbitrate--to mitigate the 
consequences of an inappropriate review.

I gladly welcome comments and suggestions aimed at broadening the constituency
of NIPS without lowering the meeting's quality.  I will summarize the feedback
I receive to the net.

Cheers, 

Mike Mozer


From postma at cs.rulimburg.nl  Fri Dec 23 05:26:23 1994
From: postma at cs.rulimburg.nl (Eric Postma)
Date: Fri, 23 Dec 94 11:26:23 +0100
Subject: Paper Announcement: Priming and Memory
Message-ID: <9412231026.AA28600@bommel.cs.rulimburg.nl>



------------------------------------------------------------------------
FTP-host: ftp.cs.rulimburg.nl
FTP-file: pub/papers/postma/memory.ps.Z
------------------------------------------------------------------------

The following paper is now available:


    The Nature of Memory Representations [6 pages]

Eric O. Postma, Ernst H. Wolf, H. Jaap van den Herik  
                        and
                Patrick T. W. Hudson

   Department of Computer Science, University of Maastricht 
      P.O.Box 616, 6200 MD Maastricht, The Netherlands 

To appear in the Proceeding of the workshop on Supercomputing in Brain
Research: From Tomography to Neural Networks, HLRZ, KFA Juelich,
Germany, November 21-23, 1994. World Scientific Publishing Company.

Abstract:
This study investigates processing and storage in the brain using a
priming task. We model a priming task with a relaxation neural network
- the Coulomb Energy Network. The network's performance on a
fragment-completion task is assessed by storing a set of memories into
the network and, subsequently, testing its completion performance.  It
has been claimed that the findings of stochastic independence on
repeated fragment completion imply noninteracting memory traces. We
nevertheless model memory in a way in which all traces embodying a
single representation interact and find stochastic independence when
the memories in the network are densely packed.  Dependence between
priming fragments can be obtained, but only when the memories are
sparsely packed.  We conclude that independence on repeated fragment
completion does not necessarily imply that the underlying memory traces
are noninteracting memory traces.

------------------------------------------------------------------------
Please do not reply directly to this message 
------------------------------------------------------------------------

FTP-instructions:

unix> ftp ftp.cs.rulimburg.nl
ftp> Name: anonymous
ftp> Password: your e-mail address
ftp> pub/papers/postma
ftp> binary
ftp> get memory.ps.Z
ftp> quit
unix> uncompress memory.ps.Z
------------------------------------------------------------------------

Merry Christmas,

Eric Postma
Computer Science Department
Faculty of General Sciences
University of Limburg
PO Box 616
6200 MD  Maastricht
The Netherlands

email: postma at cs.rulimburg.nl
web  : http://www.cs.rulimburg.nl
tel  : +31 43 883493
fax  : +31 43 252392


From jon at maths.flinders.edu.au  Fri Dec 23 06:58:43 1994
From: jon at maths.flinders.edu.au (Jonathan Baxter)
Date: Fri, 23 Dec 1994 22:28:43 +1030
Subject: TR Available: Learning Internal Representations
Message-ID: <199412231158.AA05637@calvin.maths.flinders.edu.au>

FTP-host: archive.cis.ohio-state.edu
FTP-filename: /pub/neuroprose/Thesis/baxter.thesis.ps.Z

--------------------------------------------------------

The following paper is now available:

Learning Internal Representations  [112 pages]
Jonathan Baxter

This is a preliminary draft of my PhD thesis. Note that it is in the
Thesis subdirectory of the neuroprose archive. It is in the process 
of being broken into several pieces for submission to Information and
Computation, Machine Learning and next year's COLT.

I'm afraid I cannot offer hard-copies.

ABSTRACT:
Most machine learning theory and practice is concerned with learning a
single task. In this thesis it is argued that in general there is insufficient
information in a single task for a learner to generalise well and that what
is required for good generalisation is information about {\em many
similar learning tasks}. The information about similar learning tasks forms a
body of prior information that can be used to constrain the hypothesis space
of the learner and cause it to generalise better.
Typical learning scenarios in which there are many similar tasks are image
recognition and speech recognition.

After proving that learning without prior information is impossible except in
the simplest of situations, the concept of the {\em environment} of a learner 
is introduced as a probability measure over the set of learning problems the
learner might be expected to learn. It is shown how a sample from such an
environment can be used to learn a {\em representation}, or recoding of 
the input space that is appropriate for the environment. Learning a
representation can equivalently be thought of as learning the
appropriate features of the environment.

Using Haussler's statistical decision theory framework for
machine learning, rigorous bounds are derived on the sample size
required to ensure good generalisation from a representation learning process.
These bounds show that under certain circumstances learning
a representation appropriate for $n$ tasks reduces the number of examples
required of each task by a factor of $n$. It is argued that environments
such as character recognition and speech recognition fall into the category
of learning problems for which such a reduction is possible.

Once a representation is learnt it can be used to learn {\em novel} tasks
from the same environment, with the result that far fewer examples are
required of the new tasks to ensure good generalisation. Rigorous bounds are
given on the number of tasks and the number of samples from each task required
to ensure that a representation will be a good one for learning novel tasks.

All the results on representation learning are generalised to cover any form
of automated hypothesis space bias that utilises information from similar
learning problems. 

It is shown how gradient-descent based
procedures for training Artificial Neural Networks can be generalised to cover
representation learning. Two experiments using the
new procedure are performed. Both experiments fully support the
theoretical results.

The concept of the environment of a learning process
is applied to the problem of {\em vector quantization} with the result that
a {\em canonical} distortion measure for the quantization process emerges.
This distortion measure is proved to be optimal if the task is to
approximate the functions in the environment.

Finally, the results on vector quantization are reapplied to
representation learning to yield an improved error measure for learning in
classifier environments. An experiment is presented demonstrating the
improvement. 

-------------

Retrieval Intructions:

unix> ftp archive.cis.ohio-state.edu
ftp> Login: anonymous
ftp> Password: e-mail address
ftp> cd pub/neuroprose/Thesis
ftp> binary
ftp> get baxter.thesis.ps.Z
ftp> quit
unix> uncompress baxter.thesis.ps.Z
unix> lpr -s baxter.thesis.ps
---------------

Jonathan Baxter
School of Information Science and Technology,
The Flinders University of South Australia.
jon at maths.flinders.edu.au



From 100020.2727 at compuserve.com  Fri Dec 23 11:24:19 1994
From: 100020.2727 at compuserve.com (Andrew Wuensche)
Date: 23 Dec 94 11:24:19 EST
Subject: Discrete Dynamics Lab (for DOS)
Message-ID: <941223162419_100020.2727_BHL46-1@CompuServe.COM>

Discrete Dynamics Lab 
---------------------
Announcing the first release of Discrete Dynamics Lab (described below),
a program for studying discrete dynamical networks, from Cellular
Automata to random Boolean networks, including their attractor basins. 
   
   Attractor basins are objects in space-time that link network states
according to their transitions. Access to these objects provide insights
into complexity, chaos and emergent phenomena in CA. In less ordered
networks (as well as CA), attractor basins show how the network
categorises its state space far from equilibrium, and represent the
network's memory.

   The program is released as shareware. This is a beta version and
comments are welcome. Section 1-10 in the ddlab.txt file gives an
overview of the program. Appendix A gives some operating instructions
including quick-start examples, but a detailed reference is not yet
complete. An illustrated operating manual will be released in due course.
For further background on attractor basins of CA and random Boolean
networks, and their implications, refer to refs. [1,2,3,4] below. [1] is
a book, hard copy pre-prints of [1,2,3] are available on request. 

  Platform - PC-DOS 386 or higher, with VGA or SVGA graphics, maths
co-processor, mouse, extended memory recomended. Ideally a fast 486 with
8MB ram.

download instructions
---------------------
ftp the file "ddlab.zip" (416,884 bytes):

%          ftp ftp.cogs.susx.ac.uk
name:      anonymous
password:  your complete e-mail address 
ftp>       cd pub/alife/ddlab
ftp>       binary
ftp        get ddlab.zip
ftp        close

unzip "ddlab.zip" to give the following files:

ddlab.txt    a text file describing ddlab
ddlab.exe    the program
dos4dw.exe   the DOS extender, giving access to extended memory
gliders.r_s  file with "glider" rules 
smalle.fon   two font files
sserife.fon    

to run, from the directory containing these files, enter: ddlab

All questions and comments to 

Andy Wuensche 

contact address:                      Santa Fe Institute
48 Esmond Road, London W4 1JQ UK      and The University of Sussex (COGS)
tel 081 995 8893  fax 081 742 2178    wuensch at santafe.edu
100020.2727 at compuserve.com            andywu at cogs.susx.ac.uk

Discrete Dynamics Lab
---------------------
Cellular Automata - Random Boolean Networks.
(copyright (c) Andrew Wuensche 1993)
First release of the beta version Dec 1994. 

DDLab is an interactive graphics program for research into the dynamics
of finite binary networks (for a DOS-PC platform). The program is
relevant to the study of complexity, emergent phenomena and neural
networks, and implements the investigations presented in [1,2,3,4].
   
   Using a flexible user interface, a network can be set up for any
architecture between regular CA (1d or 2d with periodic boundary
conditions) on the one hand[1], and random Boolean networks (disordered
CA) on the other[2]. The latter have arbitrary connections, and rules
which may be different at each site. The neighbourhood (or
pseudo-neighbourhood) size may be set from 1 to 9, and the network may
have a mix of neighbourhood sizes.
   
   The program iterates the network forward to display space-time
patterns (mutations are possible "on the fly"), and also runs the network
"backwards" to generate a pattern's predecessors and reconstruct its
branching sub-tree of all ancestor patterns until all "garden of Eden"
states, the leaves of the sub-tree, have been reached. For smaller
networks, sub-trees, basins of attraction or the whole basin of
attraction field can be displayed as a directed graph or set of graphs in
real time, with many presentation options. Attractor basins may be
"sculpted" towards a desired configuration. Various statistical and
analytical measures and numerical data are made available, mostly
displayed graphically. 

   The network's parameters, and the graphics display and presentation
options, can be very flexibly set, reviewed and altered. Changes can be
made "on the fly", including mutations to rules, connections or current
state. 2d networks (including the "game of life" or any mutation thereof)
can be displayed as a space-time pattern in a 3d isometric projection.
Network parameters, states, data, and the screen image can be saved and
loaded in a variety of tailor-made file formats. 

   Statistical measures and data (mostly presented graphically) include:
lambda and Z parameters, rule-table lookup frequency and entropy, pattern
density, detailed data on sub-trees, basins and fields, garden-of-Eden
density, in-degree frequency and a scatter plot of state-space.  
   
   Learning/forgetting algorithms allow attaching/detaching sets of
states as predecessors of a given state by automatically mutating rules
or changing connections. This allows "sculpting" the basin of attraction
field to approach a desired scheme of hierarchical categorisation. 
    
References. 
-----------
Wuensche,A., and M.J.Lesser. "The Global Dynamics of Cellular Automata:
An Atlas of Basin of Attraction Fields of One-Dimensional Cellular
Automata", Santa Fe Institute Studies in the Sciences of Complexity,
Reference Vol.I, Addison-Wesley, 1992.

Wuensche.A.,"The Ghost in the Machine: Basins of Attraction of Random
Boolean Networks", in Artificial Life III, Santa Fe Institute Studies in
the Sciences of Complexity, Addison-Wesley, 1994. 
 
Wuensche.A., "Complexity in One-D Cellular Automata: Gliders, Basins of
Attraction and the Z parameter", Santa Fe Institute working paper
94-04-025, 1994.

Wuensche.A., "The Emergence of Memory; Categorisation Far From
Equilibrium", Cognitive Science Research Paper 346, University of Sussex,
1994. To appear in "Towards a Scientific Basis for Consciousness" eds SR
Hameroff, AW Kaszniak, AC Scot, MIT Press. 




From G.Schram at ET.TUDelft.NL  Fri Dec 23 11:54:58 1994
From: G.Schram at ET.TUDelft.NL (Gerard Schram)
Date: Fri, 23 Dec 1994 17:54:58 +0100
Subject: Report Announcement
Message-ID: <01HKZPN1751S004ZXG@TUDERA.ET.TUDELFT.NL>

The following report is available via anonymous ftp or Mosaic:

ftp://playfair.stanford.edu/pub/reports/hastie/dann.ps.Z

	Discriminant Adaptive Nearest Neighbor Classification

		Trevor Hastie and Robert Tibshirani

We propose an adaptive nearest neighbor rule that uses local
discriminant information to estimate an effective metric for
classification.  We also propose a method for global dimension
reduction, that combines local dimension information.  In a number of
examples, the methods demonstrate the potential for substantial
improvements over nearest neighbor classification.



---------------------*********---------------------------
Gerard Schram (g.schram at et.tudelft.nl)
Control laboratory, Department of Electrical Engineering,
Delft University of Technology
P.O.Box 5031, 2600 GA Delft, The Netherlands
Phone:   +31-15-785114
Telefax: +31-15-626738
---------------------*********---------------------------



From eric at research.NJ.NEC.com  Fri Dec 23 11:59:24 1994
From: eric at research.NJ.NEC.com (Eric B. Baum)
Date: Fri, 23 Dec 94 11:59:24 EST
Subject: NIPS, blind reviewing, and elitism
Message-ID: <9412231659.AA00986@yin>

IMO the NIPS '94 program was the strongest NIPS program I've
seen, and in fact the strongest program I've seen in any
conference of even moderate size, in the sense that I saw
no weak or uninteresting papers. (However I only visited
a fraction of the posters so I may have missed some.)

Any skeptics should wait and consult the Proceedings,
which IMO will provide a Prima Facie Case that the
refereeing procedure is not broken, and should not be
fixed (at least radically).

-------------------------------------
Eric Baum
NEC Research Institute, 4 Independence Way, Princeton NJ 08540
PHONE:(609) 951-2712, FAX:(609) 951-2482, Inet:eric at research.nj.nec.com




From schmidhu at informatik.tu-muenchen.de  Fri Dec 23 12:43:29 1994
From: schmidhu at informatik.tu-muenchen.de (Juergen Schmidhuber)
Date: Fri, 23 Dec 1994 18:43:29 +0100
Subject: No subject
Message-ID: <94Dec23.184341met.42261@papa.informatik.tu-muenchen.de>


              SEMILINEAR PREDICTABILITY MINIMIZATION 
	   PRODUCES ORIENTATION SENSITIVE EDGE DETECTORS

       Technical Report FKI-201-94 (7 pages, 0.95 Megabytes) 
             Juergen Schmidhuber       Bernhard Foltin
                     Fakultaet fuer Informatik 
                  Technische Universitaet Muenchen  
                      80290 Muenchen, Germany 

			 December 24, 1994
           
    Static real world images are processed by a computationally 
    simple  and  biologically plausible  version of  the recent 
    predictability  minimization   algorithm  for  unsupervised 
    redundancy  reduction.  Without a  teacher and  without any 
    significant pre-processing, the system automatically learns 
    to generate  orientation  sensitive  edge  detectors in the 
    first (semilinear) layer.

To obtain a copy, do:

	     unix>         ftp flop.informatik.tu-muenchen.de  
	                   (or ftp 131.159.8.35)
	     Name:         anonymous
             Password:     your email address 
	     ftp>          binary
	     ftp>          cd pub/fki
             ftp>          get fki-201-94.ps.gz  (0.222 MBytes)
	     ftp>          bye
	     unix>         gunzip fki-201-94.ps.gz
	     unix>         lpr fki-201-94.ps

Alternatively, check out
http://papa.informatik.tu-muenchen.de/mitarbeiter/schmidhu.html
If your net browser does not know gzip/gunzip (which works better
than compress/uncompress), I can mail you an uncompressed 
postscript version (as a last resort).

Merry Christmas!
Juergen Schmidhuber



From wahba at stat.wisc.edu  Fri Dec 23 22:36:19 1994
From: wahba at stat.wisc.edu (Grace Wahba)
Date: Fri, 23 Dec 94 21:36:19 -0600
Subject: SS-ANOVA for `soft' classification-anct
Message-ID: <9412240336.AA14678@hera.stat.wisc.edu>

The following report is available via anonymous ftp or Mosaic:

ftp://ftp.stat.wisc.edu/pub/wahba/exptl.ssanova.ps.gz

Smoothing Spline ANOVA for Exponential Families, With Application
to the Wisconsin Epidemiological Study of Diabetic Retinopathy

        Grace Wahba, Yuedong Wang, Chong Gu
        Ronald Klein MD, and Barbara Klein MD

Given attributes (which may be discrete or continuous)
and outcomes (Class 1 or Class 0) of a sample of instances,
we develop Smoothing Spline ANOVA methods to estimate the 
*probability* of membership in Class 1, given the attribute 
vector. These methods are suitable when outcomes 
as a function of attributes are not clear-cut, as, for 
example, occurs when estimating risk of some medical outcome, 
given various predictor variables and treatments. 
These methods are penalized log-likelihood methods and 
plots of the cross sections of the estimates are generally
fairly easy to interpret in context. 
We use the results to estimate the probability
of four-year progression of diabetic retinopathy, given the three
predictor variables glycosylated hemoglobin, duration of diabetes and 
body mass index at the study baseline, based on data 
from the Wisconsin Epidemiological Study of Diabetic Retinopathy.
We discuss methods for multiple smoothing parameter selection, 
(the bias-variance tradeoff!!), numerical methods for 
computing the estimate and Bayesian `confidence intervals' 
for the estimate. We discuss methods of informal and formal
model selection (stacked generalization!!) and some open questions.
This work provides further details for work which has previously been 
announced in NIPS-93 and elsewhere.

Other related papers in the same directory include 
gacv.ps.gz, ssanova.ps.gz ml-bib.ps and theses/ywang.thesis.README
 
 Grace Wahba wahba at stat.wisc.edu .....`think snow'


From lazzaro at CS.Berkeley.EDU  Wed Dec 28 18:11:33 1994
From: lazzaro at CS.Berkeley.EDU (John Lazzaro)
Date: Wed, 28 Dec 1994 15:11:33 -0800
Subject: No subject
Message-ID: <199412282311.PAA15382@snap.CS.Berkeley.EDU>



	I sent these comments to Mike Mozer in response to his call
for NIPS comments here last week, and he suggested I pass my comments
on to the mailing list.

	As a EE, I tend to look towards IEEE conferences as examples:
there seems to be two major types.

[1] Conferences whose prime goal is to bring together as large a segment of
a research community as possible, while still maintaining a standard of
quality for presentations. IEEE Circuits and Systems and IEEE Acoustics,
Speech, and Signal Processing are two prime examples. Among other 
attributes, its expected that very many if not the majority of the 
attendees will be presenting, and that the conference's primary clientele is
the research community (both academic and industrial), as opposed to the
industrial developer.

[2] Conferences whose prime goal is to showcase the state of the art in
a field, primarily for the benefit of applied development attendees. In 
these conference, a large majority of the attendees will not be presenting
or even doing research, but are either applied developers or non-technical
observers. IEEE Solid State Circuits conference is probably the best 
example of this type of conference: it has a session on the N best new 
microprocessor designs, the N best new DRAM designs,ect. For some types of
chips, academic groups can be competitive with industry groups, and the
session is mixed; in others, a $100 million dollar investment is needed to
design the chip, and so industrial groups dominate.

        Many [1] conferences have explicit rules in submission to ensure as
many different research groups around the world are presenting
as possible: some limit the number of papers a single author can submit,
others use "membership" techniques to try to limit the number of papers from
a lab. On the other hand, [2] conferences are concerned with fairness and with
having the highest quality, but explicitly do not have "broadness" in their
charter: if the N best new Op-Amps all come from a certain company, so be it.

        NIPS has always seemed squarely in the middle of these two types
of conferences, with the "elite" aspirations of the Solid State Circuits
conference, but with a ratio of "researcher" to "developer" attendees that
is closer to the first type of conference. Certain other conferences (most
notably SIGGRAPH) started out where NIPS has been, and ended up as a [2]
type of conference -- even if you discount the majority of the SIGGRAPH
attendees who are there for the trade show and arty stuff, most attendees
of the conference papers are there to listen and learn new research ideas
for possible use in development, not to present papers themselves.      

        The tone of most of the postings on this thread seem to imply that
[1] is the direction that they'd like NIPS to move to. If so, the IEEE
experience has been that more direct approaches than "blind reviewing" can
be used to broaden the conference.

        Personally, I'd prefer [2], because I believe the technologies 
associated with NIPS are going to have the same degree of engineering 
impact as SIGGRAPH (Computer Graphics) and ISSCC (Integrated Circuits) has
had on the world, and part of realizing that impact is having a conference
of type [2]. But for this to occur, NIPS needs to market itself to the
product development community, so that the ratio of developer:researcher
at NIPS increases significantly.