X-Message-Number: 26964
From: 
Date: Thu, 8 Sep 2005 02:44:51 EDT
Subject: Sequential and parallel processing

This is more to Donaldson, but other could find if they agree or not. For  

me, there is no magic in parallel processing, our brain make a lot of sequential
 work along dendrites and from neuron to neuron. The basic function of a 

neural  network is sequential. Now, given the slowness of electrochemical 
signals, 
a lot  of parallelism must be used in the brain.
 
There is hardly any information processing in the brain with time constant  

smaller than one millisecond. Current electronics processors have a clock cycle
 smaller than one nanosecond, one million time faster. A single material  
electronics unit can so process one million more informations in a given time  

that the biological electrochemical system. For a given problem, that shift the
sequential-parallel mix on the sequential side. That is not to say that 

actual  electronics chips are fast enough to do everything on a single 
sequential  
processor.
 
When I have told that a chip could contain one thousand neurons (this is  

parallel processing), Mr. Donaldson insisted that there must be only one neuron
per chip, so he advocated an extreme form of sequential processing...
 
In fact, a brain simulation is a big problem but not even  what  
mathematicians call a hard one, such the salesman traveler. Anyway, the  

sequential-parallel option is something having a meaning only for general  
purpose computers. 
As I have said many time, only the slowest currents, varying  one the tens of 
seconds scale would be processed by such devices. Anything  faster would be 
taken into account by FPGAs (where you hardly define sequential  and parallel 
processing) or specialised neuromorphics chips mixing numerical and  analog 

computing. Clearly, the opinion of Mr. Donaldson is purely theoretical  and has 
no 
relevance to practical desing at least in the uploading frame.
 
YB.
 


 Content-Type: text/html; charset="US-ASCII"

[ AUTOMATICALLY SKIPPING HTML ENCODING! ] 

Rate This Message: http://www.cryonet.org/cgi-bin/rate.cgi?msg=26964