X-Message-Number: 25984 Date: Tue, 05 Apr 2005 18:41:58 -0400 From: Daniel Crevier <> Subject: uploading C'est un message de format MIME en plusieurs parties. Content-type: text/plain; charset=iso-8859-1 Content-transfer-encoding: 7BIT Subject: uploading For Thomas Donaldson: In message #25967 you state that "parallel computers can do things impossible for any single computer." and also that "It's simply not valid to believe that ANY parallel computer can be squeezed into a smaller number of processors than the number it uses." Can you substantitate this? I agree that parallel computers can be much faster than serial machines, but you seem to indicate that they can obtain qualitative results that serial machines cannot get. Can you give examples of this or, failing that, valid theoretical reasons why this should be so? You also state that it would be impractical to add connections between neurons in a computer simulation. It seems to me that this could be done simply by updating the matrix describing the connexions. Why should that be a problem? For Yvan Bozzonetti: A word of encouragement. I admire your attempts at modelling neurons using differential equations. This is certainly a valid approach for gaining an understanding of the detailed dynamics of neurons, but I suspect it won't be necessary for the purpose of simulating a brain. Consider an electric power system: it is comprised of thousands of transmission lines, each one a very complex system in itself, which behaves according to Maxwell's equations. Simulating the detailed currents, voltages and magnetic fields in and around a transmission line involves solving partial differential equations of quantities that propagate at the speed of light. Depending on the level of detail required, simulations involving a short length of line and a duration measured in milliseconds may take hours of number crunching with a very powerful computer. From this it would seem that calculating the flow of power in a system the size of the north american power grid is an impossible task. Well, it isn't, and you can do it on your PC! The reason is that, once you've decided that you're not interested in everything that goes on in and around transmission lines, and that all you want to know is how they *** interact with each other *** in a 60 hz AC environment, your mathematical model can be drastically simplified. That's just one example, but I believe it is typical. As one who has spent a good deal of his professional time modelling physical systems, I have found that the process tends to follow four stages: - detailed modelling: that's Maxwell's equations for power lines, and other kinds of differential equations for neurons - recoil in awe: that's the psychological reaction after you've done a bit of detailed modelling - simplification: that's when you extract the regularities and modes of behavior of the system that are relevant to the problem you're trying to solve - eureka!. That's when you put your simplifications together, and eventually build a feasible simulation that reproduces the aspects of system behaviour that are important to you. From Yvan's descriptions, and from what I understand about current efforts in neural modelling, we seem to be at the "recoil in awe" stage right now. Hopefully this won't last too long. Daniel Crevier, Ph.D. Content-type: text/html; charset=iso-8859-1 [ AUTOMATICALLY SKIPPING HTML ENCODING! ] Rate This Message: http://www.cryonet.org/cgi-bin/rate.cgi?msg=25984