From: James Rogers (jamesr@best.com)
Date: Mon Sep 08 2003 - 14:58:39 MDT
Robin Lee Powell wrote:
>
> I wonder very much what they are trying to accomplish.
>
> As far as I know, no map of the human brain down to
> individual neuron connections and weightings exists, so they
> can't be trying to mimic it in that sense.
Big caveat: the website says almost nothing about the technological details, so
this is purely speculation and maybe he has something slick up his sleeve. If
not vapor, at the very least it seems like a retread of well-trod areas. Given
the current state of knowledge re:NN this seems like an exercise in futility or
perhaps naivete, though I'm sure there are plenty of people who will disagree.
If the fellow can't explain why his flavor of the month will work on a small
scale, it seems premature to attempt it on a large scale.
Several other non-commercial and commercial research groups have tried large
neural networks on PC clusters (and with much less lofty goals), with pretty
mediocre results. The general analysis has generally been that the cluster
computing environment is badly matched to the algorithm space of neural
networks, and you don't get anywhere near the scaling factor that you want to
justify the cluster even if you make sub-optimal design changes to your
theoretical model.
Cheers,
-James Rogers
jamesr@best.com
This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:42 MDT