Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!meyering
From: meyering@cs.utexas.edu (Jim Meyering)
Newsgroups: comp.ai.neural-nets
Subject: Re: min hidden units for FF parity network?
Message-ID: <1125@ai.cs.utexas.edu>
Date: 13 Dec 90 13:58:21 GMT
References: <1122@ai.cs.utexas.edu> <1990Dec12.211633.28627@portia.Stanford.EDU>
Organization: U of TX at Austin CS Dept
Lines: 15

In article <1990Dec12.211633.28627@portia.Stanford.EDU> kortge@elaine3.stanford.edu (Chris Kortge) writes:
    Sontag shows that with sigmoid hidden units N-bit parity can be learned
    with (N+1)/2 hidden units (rounded up to the next integer).  I saw this
    in a tech report from the Rutgers U. Dept. of Mathematics titled "On
    the Recognition Capabilities of Feedforward Nets".  As I remember it
    didn't prove that this was a minimum, but it was shown to be sufficient.

    Chris Kortge
    kortge@psych.stanford.edu

For anyone else who's interested, I'm told there's also
E. Sontag "Sigmoids distinguish better than heavisides" Neural
Computation 1 (1989) 470-472.
-- 
Jim Meyering          meyering@cs.utexas.edu