Re: computer-go: Neural networks

["Richard Brown" <rbrown@xxxxxxxx>]

Julian Churchill wrote:
> The actual idea is to use a neural network trained on some professional
> games, as a sort of ultra fast pattern matcher, but it would also pick up

Yes, interesting idea...  You said half of a pair of my favorite words:

"Pattern recognition".

Pattern recognition is at the heart of AI, and will be at the core of
the best go programs of the future.  I predict that the best go programs
of the future will implement a very sophisticated sort of pattern
recognition, perhaps using neural nets as Julian describes, perhaps
using data mining, perhaps a syntactic approach, but certainly they
will recognize patterns.

Not just visual patterns, but patterns of behavior, and even _style_.

Nicol Schraudolph, Peter Dayan, and Terrence J. Sejnowski of the
Computational Neurobiology Lab at the Salk Institute wrote a paper
about neural networks which learn to evaluate 9x9 go positions using
temporal differences:
  <http://satirist.org/learn-game/systems/go-net.html>

If patterns of speech, handwriting, fingerprints, DNA sequences, and
even "heavy quark events" in high-energy physics can be recognized by
neural networks and other pattern-recognition techniques, then go too
may ultimately succumb to this approach.

The following article is a comprehensive review of pattern recognition,
and I heartily recommend it to all readers on this list:
  <http://ieeexplore.ieee.org/iel5/34/17859/00824819.pdf>

[The article is a little more user-friendly than a book I have
recommended here before:  "A Probabilistic Theory of Pattern
Recognition":  <http://www-cgrl.cs.mcgill.ca/~luc/pattrec.html> .]

Think about go programming when you read it; I hope you will enjoy it as
much as I have.  The bibliography is wonderful, too!  Here's the
abstract:

Statistical pattern recognition: a review 
- Jain, A.K.; Duin, P.W.; Jianchang Mao 
Dept. of Comput. Sci. & Eng., Michigan State Univ., East Lansing, MI,
USA
IEEE Transactions on Pattern Analysis and Machine Intelligence
               On page(s): 4 - 37 
               Jan. 2000 
               Volume: 22 Issue: 1 
               ISSN: 0162-8828 
               References Cited: 168 
               CODEN: ITPIDJ 
               INSPEC Accession Number: 6525225 

Abstract: 
The primary goal of pattern recognition is supervised or unsupervised
classification. Among the various frameworks in which pattern
recognition has been traditionally formulated, the statistical approach
has been most intensively studied and used in practice. More recently,
neural network techniques and methods imported from statistical
learning theory have been receiving increasing attention. The design of
a recognition system requires careful attention to the following issues:
definition of pattern classes, sensing environment, pattern
representation, feature extraction and selection, cluster analysis,
classifier design and learning, selection of training and test samples,
and performance evaluation. In spite of almost 50 years of research and
development in this field, the general problem of recognizing complex
patterns with arbitrary orientation, location, and scale remains
unsolved. New and emerging applications, such as data mining, web
searching, retrieval of multimedia data, face recognition, and cursive
handwriting recognition, require robust and efficient pattern
recognition
techniques. The objective of this review paper is to summarize and
compare some of the well-known methods used in various stages of a
pattern recognition system and identify research topics and applications
which are at the forefront of this exciting and challenging field.

Enjoy!

Oh, wait, here's another URL, if you'r interested in learning more about
pattern recognition.
  <http://jeff.cs.mcgill.ca/~godfried/teaching/pr-web.html>

-- 
Richard L. Brown       Office of Information Services
Unix Guy               University of Wisconsin System Administration
                       780 Regent St., Rm. 246
rbrown@xxxxxxxxxxxxxxxxx        Madison, WI  53715