Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!tut.cis.ohio-state.edu!rutgers!cmcl2!yale!Krulwich-Bruce
From: Krulwich-Bruce@cs.yale.edu (Bruce Krulwich)
Newsgroups: comp.ai.neural-nets
Subject: Re: Processing Power Equivalents
Message-ID: <51181@yale-celray.yale.UUCP>
Date: 20 Feb 89 18:18:56 GMT
References: <316@sagpd1.UUCP>
Sender: root@yale.UUCP
Reply-To: Krulwich-Bruce@cs.yale.edu (Bruce Krulwich)
Organization: Computer Science, Yale University, New Haven, CT 06520-2158
Lines: 35
In-reply-to: robinson@sagpd1.UUCP (Rob Robinson)

In article <316@sagpd1.UUCP>, robinson@sagpd1 (Rob Robinson) writes:
>Suppose the brain actually has 10 billion neurons, each having
>an average of 1000 connections with other neurons. Also assume
>that the average firing rate is 100/second.
>
>10^10 * 10^3 * 10^2 = 10^15
>		     -------
>Now imagine a system having 1 million neural type units, each having
>only 10 connections to neighboring units. Using available technology,
>assume a firing speed of say 100 Mhz.
>
>10^6 * 10 * 10^8 = 10^15
>		  -------
>Would the two systems be "equivalent" as far as processing power is
>concerned?

Depends what you mean by "processing power."  What you're saying seems to be
that the number of neuron firings (transmission across synapses) would be the
same for the two.

However, the problem is in the way the information in the net is used.  The
way most neural net models are set up, the amount of information it could
process would drop by 4 orders of magnitude (the number of neurons) and the
amount of information it could remember would drop by a 6 orders of magnitude
(the number of synapses).

Even if you quibble with my descriptions of the uses of neurons and synapses,
it is clear that the information content would decrease by many orders of
magnitude.


Bruce Krulwich
krulwich@cs.yale.edu