Evolutionary design and statistical assessment of strategies in an adversarial domain

Adversarial decision making is aimed at finding strategies for dealing with an adversary who observes our decisions and tries to learn our behaviour pattern. This contribution extends a simple mathematical model with strategies that vary along time, and motivates the use of heuristic search procedures to address the problem of finding good strategies within this new search space. The evaluation of this new class of strategies requires running a stochastic simulation so the comparison of strategies should be properly addressed. A new statistics-based technique for the comparison of strategies is also proposed and tested in this context when coupled with a Genetic Algorithm. Computational experiments showed that the new strategies are better than previous ones, and that the results obtained with this new comparison technique are encouraging.