Path: utzoo!attcan!utgpu!jarvis.csri.toronto.edu!mailrus!ames!ncar!boulder!sunybcs!sybil!jagota
From: jagota@sybil.cs.buffalo.edu (Arun Jagota)
Newsgroups: comp.ai.neural-nets
Subject: Re: : Step Function
Keywords: learning,generalization
Message-ID: <10167@eerie.acsu.Buffalo.EDU>
Date: 12 Sep 89 14:40:04 GMT
References: <1060@rex.cs.tulane.edu> <6980@sdcsvax.UCSD.Edu> <17538@bellcore.bellcore.com> <1727@cbnewsl.ATT.COM> <7011@sdcsvax.UCSD.Edu> <11308@boulder.Colorado.EDU>
Sender: nobody@acsu.buffalo.edu
Reply-To: jagota@sybil.UUCP (Arun Jagota)
Organization: SUNY @ Buffalo
Lines: 20

In article <11308@boulder.Colorado.EDU> bill@synapse.Colorado.EDU (Bill Skaggs) writes:
>
>  Is it necessary to have a bias in order to be able to learn?
>
>  Not always.  If the training set includes an example of every
>possible input, then the device only needs to be able to "memorize"
>the correct responses -- it doesn't need a preset bias.
>

Even then, it might be *useful* to consider the complexity of the 
representation (the functional form) of the learned function that the 
device forms. A device will exhibit good functional form learning ability 
if it forms a *good* representation.
Since what constitutes the *best* representation (min # of params, 
length of representation in bits, etc) is a subjective matter, this 
amounts to needing a bias. Moreover, an unbiased representation will be 
a table look-up which has no generalisation capability (whereas a 
*better* representation should have some generalisation capability).
CSNET  :  jagota@cs.buffalo.edu
BITNET :  jagota@sunybcs