New paper on Bayesian backprop

[dhw@santafe.edu]

	*** DO NOT FORWARD TO OTHER GROUPS ***



The following paper has been placed in an FTP repository at the Santa
Fe Institute. An abbreviated version of this paper will appear in the
proceedings of NIPS '93.

The paper consists of two files. Retrieval instructions appear at the
end of this message.







Bayesian Backpropagation Over I-O Functions Rather Than Weights


			David H. Wolpert

The Santa Fe Institute, 1660 Old Pecos Trail, Santa Fe, NM 87501
(dhw at santafe.edu)



Abstract: The conventional Bayesian justification for backprop is that
it finds the MAP weight vector. As this paper shows, to find the MAP
i-o function instead, one must add a correction term to backprop. That
term biases one towards i-o functions with small description lengths,
and in particular favors (some kinds of) feature-selection, pruning,
and weight-sharing. This can be viewed as an a priori argument in
favor of those techniques.







To retrieve the paper:

unix> ftp ftp.santafe.edu
Name: anonymous
Password: (Your e-mail address)
ftp> binary
ftp> cd pub/Users/dhw
ftp> get nips.93.figs.ps.Z
ftp> get nips.93.text.ps.Z
ftp> quit
unix> uncompress nips.93.figs.ps.Z
unix> uncompress nips.93.text.ps.Z
unix> lpr nips.93.figs.ps 	(or however you print postscript)
unix> lpr nips.93.text.ps	(or however you print postscript)

Note: The .figs file uncompresses to close to 2.5 meg. It may be
necessary to use the -s option to lpr to print it.