A reorganization scheme for a hierarchical system of learning automata

Abstract

Many problems in adaptive control, pattern recognition, filtering, identification, and artificial intelligence can be viewed as adaptive parameter optimization problems. The learning automaton approach to these problems has distinct advantages over the classic hillclimbing methods but suffers from high dimensionality. A hierarchical system of learning automata has been used to reduce this problem somewhat, but inefficiencies still remain, since no one hierarchical structure is optimal for the entire learning automaton operation. To resolve this problem, a reorganization scheme is introduced that uses inherit properties of ϵ-optimal learning automata to heuristically select hierarchical structures with minimal computational effort while maintaining equivalency. Simulation results demonstrate a significant reduction in convergence time when the reorganization scheme is used.