From: Ben Goertzel (ben@goertzel.org)
Date: Tue May 07 2002 - 10:28:09 MDT
We have been talking a lot about the "cognition" aspects of seed AI
The "code modification" aspects are also interesting and important
So far as I know, the most advanced work being done on automated code
modification, to date, is the work of the Supercompilers team.
This work is purely aimed, at this point, at making programs *faster* at
doing the exact same thing they already do. But similar techniques may be
used in the future to make programs do "roughly" the same thing they
currently do, faster or using less memory.
This article describes where we're at now with supercompilation:
www.goertzel.org/papers/SupercompilingJavaMay2002.htm
The links given in this article to the www.supercompilers.com website don't
work; they will work when the new supercompilers site goes up in the not too
distant future.
This article focuses on some simple examples that are easy to explain; there
is a lot more complex stuff going on experimentally, e.g. we're
supercompiling a ray tracer and some language interpreters, but there are
still bugs...
I note that my role in this research is peripheral; Val Turchin, Andrei
Klimov and Arkady Klimov are the gurus...
What I'd really like to do for these blokes is help them get funded again!
Any of y'all who work for companies that have inefficient Java code, send me
an e-mail. Maybe the supercompiler is applicable to your problem, or maybe
you just need some Russian supergenius Java program optimization experts to
do some supercompilation by hand ;> [Sales pitch over...]
-- Ben G
This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:38 MDT