Introducing a genetic generalization pressure to the Anticipatory Classifier System - part 1: theoretical approach

摘要: The Anticipatory Classifier System (ACS) is a learning classifier system that based on the cognitive mechanism of anticipatory behavioral control. Besides common reward learning, ACS able to learn latently (i.e. without getting any reward) which not possible with reinforcement techniques. Furthermore, it forms complete internal representation environment and thus use processes such as reasoning planning. Latest research observed generating accurate, maximally general rules reliably are accurate also possible), but sometimes over-specialized rules. This paper shows how genetic algorithm can be used overcome this present pressure over-specialization in generalization pressure. works then hybrid learns latently, map, evolves