Contents of Neurocomputing 21 (1998)

[Georg Thimm]

Dear reader,

Please find below a compilation of the contents for Neurocomputing and Scanning
the Issue written by V. David Snchez A.  More information on the journal are
available at the URL http://www.elsevier.nl/locate/jnlnr/05301 .

The contents of this and other journals published by Elsevier are distributed
also by the ContentsDirect service (see at the URL http://www.elsevier.nl/locate/ContentsDirect).

Please feel free to redistribute this message. My apologies if this message
is inappropriate for this mailing list; I would appreciate a feedback.

With kindest regards,

     Georg Thimm

Dr. Georg Thimm
Research scientist &                         WWW: http://www.idiap.ch/~thimm
Current Events Editor of Neurocomputing      Tel.: ++41 27 721 77 39 (Fax: 12)
IDIAP / C.P. 592 / 1920 Martigny / Suisse    E-mail: thimm at idiap.ch

********************************************************************************
Journal : NEUROCOMPUTING
ISSN : 0925-2312
Vol./Iss. : 21 / 1-3

The self-organizing map
Kohonen , Teuvo
pp.: 1-6

TEXSOM: Texture segmentation using self-organizing maps
Ruiz-del-Solar , Javier
pp.: 7-18

Self-organizing maps of symbol strings
Kohonen , Teuvo
pp.: 19-30

SOM accelerator system
Ru"ping , S.
pp.: 31-50

Sufficient conditions for self-organisation in the
one-dimensional SOM with a reduced width neighbourhood
Flanagan , John A.
pp.: 51-60

Text classification with self-organizing maps: Some lessons
learned
Merkl , Dieter
pp.: 61-77

Local linear correlation analysis with the SOM
Piras , Antonio
pp.: 79-90

Applications of the growing self-organizing map
Villmann , Th.
pp.: 91-100

WEBSOM -- Self-organizing maps of document collections
Kaski , Samuel
pp.: 101-117

Theoretical aspects of the SOM algorithm
Cottrell , M.
pp.: 119-138

Self-organization and segmentation in a laterally connected
orientation map of spiking neurons
Choe , Yoonsuck
pp.: 139-158

Neural detection of QAM signal with strongly nonlinear
receiver
Raivio , Kimmo
pp.: 159-171

Self-organizing maps: Generalizations and new optimization
techniques
Graepel , Thore
pp.: 173-190

Predicting bankruptcies with the self-organizing map
Kiviluoto , Kimmo
pp.: 191-201

Developments of the generative topographic mapping
Bishop , Christopher M.
pp.: 203-224


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

			    Neurocomputing 21 (1998)
			     Scanning the issue 

T. Kohonen presents  in The self-organizing map  (SOM) as overview  of the SOM
algorithm which  realizes a data-compressing mapping  of a high-dimensional
data distribution onto a  regular lower-dimensional grid. The SOM preserves
the  most important topological  and metric  relationships of  the original
data, this leads to its main properties of visualization and abstraction.

In    TEXSOM:    Texture    segmentation   using    self-organizing    maps
J. Ruiz-del-Solar  describes a  texture segmentation architecture  based on
the   joint  spatial/spatial-frequency   paradigm.   The  Adaptive-Subspace
Self-Organizing Map  (ASSOM) or the  Supervised ASSOM (SASSOM) are  used to
generate  automatically  the  oriented  filters. Defect  identification  on
textured images can be performed using the proposed architecture.

T.  Kohonen  and  P.   Somervuo  present  Self-organizing  maps  of  symbol
strings.  Instead of  defining metric  vector spaces  to be  used  with the
self-organizing  map (SOM)  as is  usually  the case,  symbols strings  are
organized on a SOM array. The  batch map principle and average over strings
are applied. The  Feature Distance (FD) between the  symbol strings and the
Redundant Hash Addressing (RHA) are  employed for the construction of large
SOM arrays.

S.  Rping,  M. Porrmann,  and  U.  Rckert  describe the  SOM  accelerator
system. The system is a massively  parallel system based on the NBISOM - 25
chips.  Each  chip  contains  an   array  of  five  times  five  processing
elements. A  VMEBus board  with sixteen  chips on it  was built.  The model
vectors have up  to 64 weights with 8-Bit  accuracy. The system performance
is 2.4 GCUPS for learning and 4.1 GCPS for recall.

J.A. Flanagan  presents Sufficient conditions for  self-organisation in the
one-dimensional  SOM   with  a  reduced  width   neighborhood.  To  achieve
self-organization three  intervals of  non-zero probability separated  by a
minimum distance are required. The conditions are sufficient, but appear to
be close to necessary.

D. Merkl  describes in Text classification with  self-organizing maps: Some
lessons learned  the application of a hierarchical  neural network approach
to the  task of document  classification. The different network  levels are
realized by independent SOMs allowing  the selection of different levels of
granularity while exploring the document collection.

A. Piras and  A. Germond present in Local  linear correlation analysis with
the  SOM an extension  to the  SOM algorithm  for selecting  relevant input
variables in nonlinear regression. The linear correlation between variables
in  neighbor spaces  is  studied  using the  SOM.  A localized  correlation
coefficient  is determined  that allows  the understanding  of  the varying
dependencies over the definition manifold.

In  Applications  of  the  growing  self-organizing  map  T.  Villmann  and
H.-U. Bauer describe the growing self-organizing map (GSOM) algorithm. This
extension of the SOM algorithm adapts  the topology of the map output space
and  allows for unsupervised  generation of  dimension-reducing projections
with optimal neighborhood preservation.

In  WEBSOM  -  Self-organizing  maps  of  document  collections  S.  Kaski,
T. Honkela,  K. Lagus,  and T.  Kohonen describe a  new method  to organize
large collections of full-text documents in electronic form. A histogram of
word categories is used to  encode each document. The documents are ordered
in a  document map according  to their contents.  Computationally efficient
SOM algorithms were used.

In Theoretical  aspects of  the SOM algorithm  M. Cottrell, J.C.  Fort, and
G.  Pges review  the  status quo  of  the efforts  to  understand the  SOM
algorithm theoretically. The study  of the one-dimensional case is complete
with the  exception of  finding the appropriate  decreasing rate  to ensure
ordering. The study of the multi-dimensional case is difficult and far from
being complete.

Y. Choe and R. Miikkulainen present Self-organization and segmentation in a
laterally connected orientation map  of spiking neurons. The model achieves
selforganization   based   on  rudimentary   visual   input  and   develops
orientation-selective  receptive fields  and patterned  lateral connections
forming  a global  orientation map.  Crucial principles  of  the biological
visual cortex are incorporated.

In  Neural  detection  of  QAM  signal  with  strongly  nonlinear  receiver
K. Raivio, J. Henriksson, and O. Simula describe neural receiver structures
for adaptive  discrete-signal detection. A receiver structure  based on the
self-organizing  map  (SOM)  is  compared with  the  conventional  Decision
Feedback  Equalizer  (DFE)  for  a  Quadrature  Amplitude  Modulated  (QAM)
transmitted signal.

T. Graepel,  M. Burger and  K. Obermayer describe in  Self-organizing maps:
Generalizations  and  new  optimization  techniques  three  algorithms  for
generating topographic  mappings based on cost  function minimization using
an EM  algorithm and deterministic annealing. The  algorithms described are
the Soft Topographic Vector Quantization (STVQ) algorithm, the Kernel-based
Soft Topographic Mapping (STMK) algorithm, and the Soft Topographic Mapping
for Proximity data (STMP) algorithm.

K.  Kiviluoto  presents Predicting  bankruptcies  with the  self-organizing
map. The qualitative analysis on  going bankrupt is performed using the SOM
algorithm in  a supervised manner.  The quantitative analysis  is performed
using three classifiers: SOM-1, SOM-2, and RBF-SOM. Results are compared to
Linear Discriminant Analysis (LDA) and Learning Vector Quantization (LVQ).

C.M. Bishop, M.  Svensn, and C.K.I. Williams present  Developments of the
generative  topographic  mapping  (GTM)  which  is an  enhancement  of  the
standard  SOM algorithm  with a  number  of advantages.  Extensions of  the
GTMare reported: the  use of an incremental EM algorithm,  the use of local
subspace  models, mixing  discrete  and  continuous data,  and  the use  of
semi-linear models and high-dimensional manifolds.

I  appreciate the cooperation  of all  those who  submitted their  work for
inclusion in this issue.

V. David Snchez A.

Editor-in-Chief