Variable Hierarchical Structure Learning Automata with a Stationary Random Environment at Each Level

Abstract

In this paper, the learning problem in which the details of learning objects are not known clearly in advance is considered. For this problem, a new hierarchial structure learning automata (variable hierarchical structure learning automata) is proposed, which has a stationary random environment at each level. This automata changes its structure based on the new information of learning object. The learning property of this automata is considered theoretically and it is proved that the probability of this automata finding the best object can be approached 1 as much as possible. And, by numerical simulation, the profile of learning by this algorithm is shown.