Conditional Optimality of Learning Automata with 2-state Bayesian Estimators

抄録

A novel learning automaton with 2 state Bayesian estimators, which is based on β-type learning automaton, is considered. The β-type learning automaton with multiple Bayesian estimators is our original work and the scheme of it has been proven to be conditionally optimal under some environments. Furthermore it was shown that β-type learning automaton exhibits superior learning behavior compared to conventional learning automata through several experimental results. However, the β-type one is inferior from the viewpoints of memory usage and computational efficiency. So, the β-type learning automaton which consists of 2-state Bayesian estimators as minimum resources has been proposed in our previous work. In this study, we show the sufficient condition for conditional optimality of the β-type LA which consists of 2-state Bayesian estimators.