Strategy Selection by Genetic Programming
Gernot Stenz, Andreas M. Wolf · The Florida AI Research Society · 1999
Strategy parallelism is a powerful concept for applying parMicUsm to automated theorem proving. One of the most important problems to be solved in this approach is the proper distribution of the available resources mnong the different strategies. This task usually requires a lot of user expertise. When the resource distribution has to be done automatically, an adaptive algorithm must be used to optimize prover performance. We introduce a genetic algorithm that can be used for such an optimization and we show lmw such an algorithm can be integrated with other methods for automatic prover configuratkm. Ore give some experimental data to veri~, the validity of our approach and explain g~me of the future development possibilities.