Improving Performance in Size-Constrained Extended Classifier Systems

摘要

Extended Classifier Systems, or XCS, have been shown to be successful at developing accurate, complete and compact mappings of a problem's payoff landscape. However, the experimental results presented in the literature frequently utilize population sizes significantly larger than the size of the search space. This resource requirement may limit the range of problem/hardware combinations to which XCS can be applied. In this paper two sets of modifications are presented that are shown to improve performance in small size-constrained 6-Multiplexer and Woods-2 problems.