---------- Forwarded message ----------
Date: Wed, 25 Jun 97 16:55:45 JST
From: Hugo de Garis <degaris@hip.atr.co.jp>
To: gann-list@cs.iastate.edu
Subject: GANN: MAXIMIZING EVOLUTION SPEED & CAM-BRAIN MACHINE : Hugo de Garis, ATR
Dear GANNers,
My hardware designer colleague Dr. Michael Korkin and I are about to debut
the CAM-Brain Machine (CBM) prototype (at the GP97 conf) which is an FPGA
(Xilinx XC6264 chip) based piece of hardware which updates 3D Cellular
Automata (CA) cells at the incredible rate of over 100 Billion (with a B) a
second, which makes the evolution of CA based neural network modules possible
in less than a second (i.e. a complete run of a GA, with tens of thousands of
CA based neural net circuit growths and fitness evaluations). If this machine
is as successful as we hope it will be, then I believe the field of GAs and
particularly its offshoot "Evolutionary Engineering" will be radically
changed. It should be possible to evolve tens of thousands of such modules to
be assembled into humanly specified "artificial brain" architectures. If
"Brain Building" takes off, then large numbers of people will be involved in
evolving large numbers of modules for all kinds of artificial brain controlled
products, which in turn will emphasize the need for maximizing the speed of
evolution. When one looks in the GA literature, there is not a lot on
techniques for evolving AS FAST AS POSSIBLE. I appeal to the GA research
community to devote more effort into MAXIMIZING EVOLUTION SPEEDS, and I dont
just mean in terms of hardware, but more in terms of methodologies.
Cheers,
Dr. Hugo de Garis.
http://www.hip.atr.co.jp/~degaris
http://www.genobyte.com
P.S. If the first CBM is a success (so far, software simulation experiments
which test the evolvability of the CBM are looking good), then it is likely
that further CBMs will be purchasable at about $70,000 each, but probably not
in 1997, more likely in 1998.