RE: [Fwd: [>Htech] Artificial Development to Build World's Biggest Spiking Neural Network]

From: AD.COM - Marcos Guillen (
Date: Tue Sep 16 2003 - 18:42:10 MDT

Thanks, Mez, we are aware of some of those past efforts. I agree that those
are samples of important work we must consider as a guidance in our
efforts, as you will not doubt agree with me that those projects show us
what NOT to do, but do not show what CAN'T be done.

Mass simulation in a computer constrained environment is a challenging task,
and we do believe this is a classical computer science challenge, not only a
neurology mystery. We think that this type of challenge is not so different
from the task before anyone trying to take on Google or SAP.

What type of computer infrastructure do you need to integrate the supply
chain, product and order fulfilment, customer support, payment orders and
banking needs of a world wide corporation?
How many ops does it take to look for a given phrase in a pool of 3,3
billion web pages, rank the results, re-rank the winners, and orderly
display those for thousands of concurrent users?

Well, certainly not us many as a candid observer would estimate. The
difference between theoretical computation taking place in a complex query,
and the real computer power needs of Google, is what makes possible its
astonishing performance, and the main reason why, so far, Microsoft has been
unable to circumvent their patents and produce the competing search engine
they promised 2 years ago.

Our job is to take over that challenge, and we have some modest reasons to
believe that we may succeed, reasons good enough to dedicate my resources,
my employees, and my full, an distracted attention to that task from now on.

But over all, we very much welcome the help, comments and frank dialog with
like minded individuals like you, what is why we are in the Valley in the
first place!

Best regards,
Marcos Guillen

>Marcos, as I said in email to you and on this list, I like your
>architecture and the fact that it allows you to easily update the
>simulation techniques you use. I'm very pleased to see someone taking
>on a project of this sort, and in particular doing so without the
>constraints of a short-term project.
>I also agree with you that we don't know yet exactly what level of
>granularity will be required to simulate the brain. My only complaint
>is that we can be fairly certain that the required level of
>granularity is quite a bit beyond what you currently estimate. How do
>we know this? Because other groups have spent years working on neural
>simulations and have found that, without getting down to well below
>single-neuron resolution, their simulations produce results that are
>wildly different from what we see in actual brains*.
>So again, I congratulate you on the audacity of trying to pull off
>this project. Your approach on the software side seems sound. I
>would only suggest that you reconsider your estimates of how far away
>you are from being able to accurately simulate a real human brain.
>* - See for example:
>1) The NEURON Neural Simulation Environment
>2) The book _Computational Neuroscience: Realistic Modeling for
>Experimentalists_, CRC Press (2000)

Best Regards,
Marcos Guillen

This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:42 MDT