CONTINUOUS ACTION SET LEARNING AUTOMATA FOR STOCHASTIC OPTIMIZATION

The problem of optimization with noisy measurements is common in many areas of engineering. The only available information is the noise corr upted value of the objective function at any chosen point in the param eter space. One well known method for solving this problem is the stoc hastic approximation procedure. In this paper we consider an adaptive random search procedure, based on the reinforcement learning paradigm. The learning model presented here generalizes the traditional model o f a learning automaton [Narendra and Thathachar, Learning Automataa: A n Introduction, Prentice Hall, Englewood Cliffs, 1989]. This procedure requires a lesser number of function evaluations at each step compare d to the stochastic approximation. The convergence properties of the a lgorithm are theoretically investigated. Simulation results are presen ted to show the efficacy of the learning method.