GP Applications

Back to main GP Brainstorming page.

Relevant papers from the Symposium Working Notes:

Bargaining in a Three-Agent Coalitions Game: An Application of Genetic Programming. Garett Dworman, Steven O. Kimbrough, and James D. Laing

Genetic Programming as a Means of Assessing and Reflecting Chaos E. Howard N. Oakley.

Applying Genetic Programming to Intrusion Detection Mark Crosbie and Eugene H. Spafford.

Predicting Whether Or Not a 60-Base DNA Sequence Contains a Centrally-Located Splice Site Using Genetic Programming. Simon Handley

Automated Discovery of Protein Motifs with Genetic Programming. John R. Koza and David Andre

Genetic Programming Controlling a Miniature Robot. Peter Nordin and Wolfgang Banzhaf

Handley wants to evolve statistic measures, which may be more difficult than evolving convolution functions. A simple "add to memory" operation (ADDMEM MEMLOC WHATTOADD) may help this goal. (Handley's paper describes another approach) {Eric Siegel}

Wolfgang [Banzhaf] showed video of 4inch diameter robot moving about an irregular pen under the control of an evolving GP population of 50 programs. Takes 20 mins to learn to move and avoid running into the walls. {William Langdon}

Perhaps only a small part of the audience working on applications? {William Langdon}

The event-driven GP used in Conor Ryan's work also was interesting - I never thought of using GP in such a way before. I am investigating ways of applying it to my research in security. {Mark Crosbie}

1 - Creating test cases, particularly in real life situations is very important. How can we show the testcases we are using a genuinely representative of the problem as a whole?

2 - Just because GP works well on our training/test data, how can we say with certainty to outsiders that it will work well on unseen data.

3 - Real life is noisy, not only should training data reflect this, but having noise may also help overfitting.

4 - What has actually been done with GP? Most of the papers (including mine, I must admit) dealt with topics of interest only to GPers. I know there were only two sessions on applications, but Mark Crosbie's paper seemed to be the only one to obviously say "Yes, here's a real problem that a lot of people care about, and GP can help". Perhaps this is being a bit unfair to Simon Handley and David Andre who had papers on applying GP to analysing DNA and proteins - but how real world is this? Like I said, maybe I am being unfair, and maybe it is of importance, but GP really needs people to be able to say, "Okay, GP works well enough for me, and here's what I did with it"

{Conor Ryan}