Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!uunet!spool.mu.edu!uwm.edu!rpi!ispd-newsserver!kodak!isctsse!gabber!rao
From: rao@gabber.kodak.com (Arun Rao)
Newsgroups: comp.ai.neural-nets
Subject: Re: continuous vs discrete values for weights
Message-ID: <1991Feb5.165813.10038@usenet@kadsma>
Date: 5 Feb 91 16:58:13 GMT
References: <1991Feb2.001242.3473@watserv1.waterloo.edu>
Sender: usenet@usenet@kadsma (News Administrator)
Reply-To: rao@gabber.kodak.com (Arun Rao)
Organization: Image Electronics Center, Eastman Kodak Company
Lines: 22

In article <1991Feb2.001242.3473@watserv1.waterloo.edu>, ssingh@watserv1.waterloo.edu (The Sanj-Machine aka Ice) writes:
... [stuff deleted ]
|> 
|> I always thought that neurons assume one of a finite set of strengths. It
|> is just that it is a very large set, so from our vantage point it
|> appears continuous.
... [stuff deleted ]

	How large is very large ?  It appears unlikely to me that neuron
	activation could possess as much resolution as (say) even a typical
	binary float representation.  I don't remember having seen any numbers,
	but I would tend to think that if you need more than 8 bits of
	resolution to get a neural computational model to work, the biological
	plausibility of such a model is suspect.

	This is not to say, of course, that biological plausibility should be
	the acid test in evaluating models, especially application-oriented work.

	I'd be glad to hear about any experimental evidence that supports a
	considerably higher resolution in individual neuron activation.

	-Arun