ASSOCIATIVE REINFORCEMENT LEARNING - FUNCTIONS IN K-DNF

An agent that must learn to act in the world by trial and error faces the reinforcement learning problem, which is quite different from stan dard concept learning. Although good algorithms exist for this problem in the general case, they are often quite inefficient and do not exhi bit generalization. One strategy is to find restricted classes of acti on policies that can be learned more efficiently. This paper pursues t hat strategy by developing algorithms that can efficiently learn actio n maps that are expressible in k-DNF. The algorithms are compared with existing methods in empirical trials and are shown to have very good performance.