Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!iuvax!rutgers!ucsd!ucbvax!decwrl!hplabs!hp-sdd!ucsdhub!sdcsvax!beowulf!pluto
From: pluto@beowulf.ucsd.edu (Mark E. P. Plutowski)
Newsgroups: comp.ai.neural-nets
Subject: Re: : Step Function
Keywords: bias in learning,generalization
Message-ID: <7024@sdcsvax.UCSD.Edu>
Date: 5 Sep 89 20:12:11 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>
Reply-To: pluto@beowulf.UCSD.EDU (Mark E. P. Plutowski)
Organization: EE/CS Dept. U.C. San Diego
Lines: 35

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.
>

This is a strong form of a learning bias.  
The device assumes that everything
it sees during training is relevant, 
and, sufficient.  

>  It may be more interesting to wonder about more "intelligent"
>devices, which are capable of learning from commentary as well as
>examples.  ("Commentary" is, e.g., your English teacher telling you
>that all proper nouns should be capitalized. ... 

You might be interested in some approaches to learning theory
in which the device has access to a teacher that can provide more
than just a error-signal  ... e.g., once a hypothesis is formed by
the device, the teacher, rather than just saying whether or not
the hypothesis is correct for given values of inputs, can provide
counterexamples to the device.  Of course, this requires a teacher
with expert knowledge!  And, the ability to ascertain what 
hypothesis the device currently entertains.  

Applying this idea to neural networks is difficult.  
The question is: how to apportion this type of training
signal to the appropriate units in the network.  And, once
received by the pertinent units, what to make of it.  
Any ideas on how to backpropate such training information?
"Commentary" feedback can apply to the entire hypothesis formed 
by the device, not just its performance on a particular input.