A Fitness Proportionate Reward Sharing: a Viable Default Hierarchy Formation Strategy in LCS

The learning task in a Learning Classifier System (LCS) is aimed at building a set of rules that work in coordination to accurately model a given environment. The addition of the hash symbol (‘#’) in LCS’s condition provides varying degree of coverage to environmental niches. Building a hierarchical set of rules, where accurate and more specific rules respond to a subset of the situations covered by more general but less accurate default rules will be vital to achieve a compact rule set size, especially when dealing with an environment that has huge numbers of states. However, the formation of viable default hierarchy in LCS has been a nightmare in this research area for decades. This paper presents a new resource allocation scheme that leads to the formation of a default hierarchy in LCS. A fitness proportionate reward sharing scheme is introduced and the performance of the algorithm is tested using known test functions.