Lambda the Ultimate

inactiveTopic USENIX JVM Symposium '02
started 8/19/2002; 5:16:28 AM - last post 8/20/2002; 10:23:42 AM
Ehud Lamm - USENIX JVM Symposium '02  blueArrow
8/19/2002; 5:16:28 AM (reads: 1667, responses: 2)
USENIX JVM Symposium '02
This was, without a doubt, one of the most interesting conferences I've ever attended. Blatantly research-oriented and academic in nature, the technical sessions were fascinating to watch.

Interesting conference report.

I am interested in the Machine Learning for Memory Management concept. I am a long time fan of adaptive software (self-tning software, if you will). I am also interested in seeing real life examples of machine learning outperforming static heuristics. This is not as common as you'd think.


Posted to cross-language-runtimes by Ehud Lamm on 8/19/02; 5:18:10 AM

Noel Welsh - Re: USENIX JVM Symposium '02  blueArrow
8/19/2002; 9:38:26 AM (reads: 496, responses: 0)

The Machine Learning for Memory Management paper is available from the author's homepage

The USENIX JVM Symposium has kindly password protected their copy.

Noel Welsh - Re: USENIX JVM Symposium '02  blueArrow
8/20/2002; 10:23:42 AM (reads: 426, responses: 0)
Further notes on the Machine Learning paper. It's an extract from what appears to be a Honours project. The paper is inaccurate in some places (e.g. the treatment of generalisation in section 4.2 is incorrect; neural networks and regression are techniques for generalisation) and maddeningly incomplete (the full experimental setup is not detailed). More analysis of the results would also be good (there are no tests for significance, for example). However the idea is still cool. The only use of AI in compilers that I'm aware of are genetic algorithms for compiler stage ordering (I think I posted a link some time ago) and this:

http://sunshine.cs.uni-dortmund.de/~knoop/PLDI2002/Abstract/72.html