preprint available (Efficient training of multilayer percpetrons using PCA)

[Michael Biehl]

Dear Colleagues,
the following preprint is now available   online at:
                   http://www.cs.rug.nl/~biehl/prepneuro.html
(pdf format, 14 pages, 5 figures)


Efficient training of multilayer perceptrons using PCA
C. Bunzmann, M. Biehl, R. Urbanczik
(14 pages, 5 figures)

                                              Abstract 
A training algorithm for multilayer perceptrons is discussed and studied
in detail, which relates to the technique of Principal Component Analysis.
The latter is performed with respect to a correlation matrix computed from 
the example inputs and their target outputs.
Typical properties of the training procedure are investigated by means of a 
statistical physics analysis in models of learning regression and 
classification tasks. 
We demonstrate that the procedure requires by far fewer examples  for good
generalization than traditional on-line training. For networks with large 
hidden layers we derive the training prescription which achieves, within our 
model,  the optimal generalization  behavior.



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

    Michael Biehl

    Rijksuniversiteit Groningen   
    Wiskunde & Informatica
    Blauwborgje 3 
    9747 AC Groningen
    The Netherlands

    e-mail biehl at cs.rug.nl
    web    www.cs.rug.nl/~biehl