Newsgroups: comp.ai
Path: cantaloupe.srv.cs.cmu.edu!das-news2.harvard.edu!news2.near.net!howland.reston.ans.net!news.sprintlink.net!noc.netcom.net!netcom.com!olea
From: olea@netcom.com (Michael Olea)
Subject: Re: Request - Learning Stratergies
Message-ID: <oleaD9qsGu.78o@netcom.com>
Organization: NETCOM On-line Communication Services (408 261-4700 guest)
References: <3qv56v$p6f@caesar.wits.ac.za>
Date: Tue, 6 Jun 1995 08:19:41 GMT
Lines: 39
Sender: olea@netcom10.netcom.com

qgregory@concave.cs.wits.ac.za (Quentin Gregory) writes:

>Hi, Matt again.
>Doea anyone have some info on the different types of learning stratergies that 
>exist, that are not implemented using neural-networks. (e.g. genetic algorithms,
> Q learning, ...)

>I would be most appreciative if you could mail or post relevent papers, 
>journals, ... where the info could be found.

>Quentin Gregory, Bsc
>Email : qgregory@cs.wits.ac.za


	There is a huge number of strategies. One way to classify them
is by the data they work on, e.g. feature vectors, variable length strings,
and graphs.  Neural-networks work on feature vectors.  All feature vector
approaches ultimately partition some feature space into equivalence classes;
learning is a matter of constructing a partition, recognition is a matter
of mapping a vector to its equivalence class.  Aside from neural nets there
are decision trees (see, e.g. J. Ross Quinlan's book "C4.5-Programs for
Machine Learning", ISBN 1-55860-238-0), nearest neighbor matching (see
"Nearest Neighbor(NN) Norms: NN Pattern Classification Techniques", B. V.
Dasarathy (ed.) ISBN 0-8186-8930-7), and extraction of Bayesian Networks
from training data (see "Probabilistic Reasoning in Intelligent Systems:
Networks of Plausible Inference", Judea Pearl ISBN 0-934613-73-7; also
look at the Proceedings of the Nth Conference on Uncertainty in Artificial
Intelligence, which usualy have some papers on the construction of Bayesian
Networks from training data) - this just scratches the surface of feature
vector approaches.  "Syntactic Pattern Recognition-An Introduction", Gonzalez
and Thomason, ISBN 0-201-02931-6, gives some methods for learning grammars
from data (it might be out of print).  "Structural Pattern Recognition",
T. Pavlidis, ISBN 3-540-08463-0 deals with graph matching schemes.  These
all deal mainly with "Supervised Learning", where there is training data
labled by class membership.  "Unsupervised Learning" treats the cases where
classes are not defined a priori - these are essentially cluster analysis
schemes (many books treat cluster analysis).  Simulated Annealing is another
"learning" scheme... Again, there is a huge array of learning strategies.

