new papers

[Juergen Schmidhuber]

The following papers are now available:

-------------------------------------------------------------------
DISCOVERING PREDICTABLE CLASSIFICATIONS 
(Technical Report CU-CS-626-92)
 ..
Jurgen Schmidhuber, University of Colorado 
                                      ..  ..
Daniel Prelinger, Technische Universitat Munchen 

ABSTRACT:    Prediction problems are among the most common learning 
problems for  neural networks  (e.g. in the context  of time series 
prediction,  control,  etc.).   With many  such problems,  however, 
perfect  prediction  is  inherently impossible.  For such  cases we 
present novel unsupervised systems that  learn to classify patterns 
such that the classifications are predictable  while still being as 
specific  as possible.  The approach  can be  related  to the  IMAX 
method  of  Hinton,  Becker  and  Zemel  (1989, 1991).  Experiments 
include  Becker's and  Hinton's  stereo task,  which can  be solved 
more readily by our system. 






-------------------------------------------------------------------
PLANNING SIMPLE TRAJECTORIES USING NEURAL SUBGOAL GENERATORS 
(for SAB92)
 ..
Jurgen Schmidhuber,  University of Colorado 
                                        ..  ..
Reiner Wahnsiedler, Technische Universitat Munchen 

ABSTRACT:   We consider the  problem of reaching a given goal state 
from a given start state by letting  an `animat' produce a sequence 
of  actions  in an  environment  with  multiple  obstacles.  Simple 
trajectory  planning  tasks are  solved with  the help  of `neural' 
gradient-based   algorithms  for   learning  without  a  teacher to 
generate  sequences of  appropriate subgoals  in response  to novel 
start/goal combinations.






-------------------------------------------------------------------
STEPS TOWARDS `SELF-REFERENTIAL' LEARNING: A THOUGHT EXPERIMENT 
(Technical Report CU-CS-627-91)
 ..
Jurgen Schmidhuber, University of Colorado 

ABSTRACT:   A major  difference between human learning and machine 
learning  is that  humans  can reflect  about their  own  learning 
behavior  and  adapt  it to  typical  learning  tasks  in a  given 
environment. To make some initial theoretical steps toward `intro-
spective' machine learning,  I present - as a thought experiment -
a `self-referential'  recurrent neural network  which can  run and 
actively  modify its  own  weight  change  algorithm.  Due to  the 
generality of the architecture, there are no theoretical limits to 
the  sophistication  of  the  modified  weight  change  algorithms 
running on the network  (except for unavoidable pre-wired time and 
storage constraints).  In theory,  the network's weight matrix can 
learn not only to change itself, but it can  also learn the way it 
changes itself,  and the  way it changes the way it changes itself  
--- and  so on  ad infinitum.  No endless  recursion is  involved, 
however.  For one variant of the  architecture, I present a simple 
but general  initial reinforcement learning algorithm. For another 
variant, I derive a  more complex  exact  gradient-based algorithm 
for supervised  sequence learning.  A  disadvantage of  the latter 
algorithm  is its  computational complexity per time step which is 
independent of  sequence length and equals O(n_conn^2 log n_conn), 
where n_conn  is the number  of connections.  Another disadvantage 
is the high number of  local minima of the unusually complex error 
surface.  The purpose of my thought experiment, however, is not to 
come up with the most  efficient or most practical `introspective' 
or `self-referential' weight  change algorithm,  but to  show that 
such algorithms are possible at all.


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

To obtain copies, do:

             unix>         ftp archive.cis.ohio-state.edu
             Name:         anonymous
             Password:     (your email address)
             ftp>          binary
             ftp>          cd pub/neuroprose
             ftp>          get schmidhuber.predclass.ps.Z
             ftp>          get schmidhuber.subgoals.ps.Z
             ftp>          get schmidhuber.selfref.ps.Z
             ftp>          bye
             unix>         uncompress schmidhuber.predclass.ps.Z
             unix>         uncompress schmidhuber.subgoals.ps.Z
             unix>         uncompress schmidhuber.selfref.ps.Z
             unix>         lpr  schmidhuber.predclass.ps
             unix>         lpr  schmidhuber.subgoals.ps
             unix>         lpr  schmidhuber.selfref.ps

---------------------------------------------------------------------
Sorry, no hardcopies. (Except maybe in very special urgent cases).
 ..
Jurgen 

      

Address until December 17, 1992:
 ..
Jurgen Schmidhuber 
Department of Computer Science 
University of Colorado 
Campus Box 430 
Boulder, CO  80309, USA 
email: yirgan at cs.colorado.edu 


Address after December 17, 1992:
 ..
Jurgen Schmidhuber 
          .. 
Institut fur Informatik
                    ..  ..
Technische Universitat Munchen 
                    ..
Arcisstr. 21, 8000 Munchen 2, Germany

email: schmidhu at informatik.tu-muenchen.de