[Coco] FPGA

[Andrew]

Alan,

You have already been given a lot good of information about FPGAs and 
such, so I won't reiterate that.

What I will mention is that if you ever want to see the tantalizing 
possibilities of FPGAs, google the terms "CAM Brain", "Hugo DeGaris", 
"roboneko", "Brain Building", "StarLabs" and "Genobyte".

For those not in the know:

Back in the mid-to-late 1990s, Hugo DeGaris came up with this scheme to 
build what was essentially a CA (cellular automata) driven, evolvable 
neural-net system built in hardware, based on a specific model of Xilinx 
FPGA (dig deep enough, and you will find out which FPGA, which is no 
longer manufactured, and why it was ideal). Basically, the system would 
evolve a neural-net, in which each FPGA represented a portion of the 3D 
logical substructure (cube shaped, from images) which contained a 
portion of the neural network, the interconnections and such being 
shaped by an evolutionary CA algorithm. The FPGA simulated the neurons 
and the network they formed.

DeGaris' idea was to build a brain using this method - his first goal 
was that of the equivalent of a cat's brain (hence, the roboneko 
project, which was a robot to simulate a cat, which the CAM Brain would 
control - it looked kinda like Sony's Aibo - but was never built, only 
modeled). Some sub-goals were met with success (I think some nets were 
evolved to discriminate sounds and the like). A few of the initial CAM 
Brain machines were built by Genobyte and shipped to a few places around 
the world, mainly ATR in Japan and StarLabs in Brussels, which is 
defunct (there are pictures - the case of these initial machines were 
designed to look like a slice of the brain cortex - very nice industrial 
design).

I remember at the time (late-1990's) that there was a lot of discussion 
on the merits of the system, what it could or couldn't do, and whether 
the whole thing was a scam or smoke and mirrors. It is still unclear 
what the verdict is. Hugo DeGaris has bounced around to other places 
since then, first to a campus in Utah, then later (and currently) to a 
university in China. He is a bit "out there" with his ideas, it may be a 
case of overworked genius or what-have-you (for instance, his ideas 
which led to his book, The Artilect War). He is a controversial figure 
in the AI community (though not shunned like Mentifex).

I don't know what happened to the machines which were built; it's a very 
strange case to dig into, and I have never found any real answers. The 
idea seems sound, but at the same time I only lurk on the edges so I 
don't know what the real truth is. I am pretty certain though that real 
hardware and development was done, but due to lack of quick progress 
(the bane of AI), the dot-COM bust, as well as a whole host of other 
reasons (not the least of which is probably DeGaris' personality, though 
that is just speculation on my part), this project fizzed out. Because 
of all of this, his work seems like a scam to get money, bouncing from 
benefactor-to-benefactor.

What the ultimate truth is remains speculation. But the idea of using 
FPGAs to evolve neural networks remains a fascinating area of research, 
at least for me (I only wish I had the time and money to learn FPGA 
design and programming, etc - but it is something that looks like it 
would take several years of intense study to really understand it to the 
point of performing my own experiments in the area, and I need another 
project like a hole in the head).

-- Andrew L. Ayers, Glendale, Arizona

> Message: 6
> Date: Wed, 18 Feb 2009 14:51:43 -0500
> From: N8WQ <exwn8jef at gmail.com>
> Subject: [Coco] FPGA
> To: CoCoList for Color Computer Enthusiasts <coco at maltedmedia.com>
> Message-ID: <499C66CF.6020001 at gmail.com>
> Content-Type: text/plain; charset=ISO-8859-1; format=flowed
> 
> Folks,
> Occasionally I hear the term FPGA used in various discussions here on 
> the list. Can anyone recommend any books, tutorials, or url's that can 
> help me learn about this subject?
> 
> Alan Jones