A New Continuous Action-Set Learning Automaton for Function Optimization

摘要

In this paper, we study an adaptive random search method based on learning automaton for solving stochastic optimization problems in which only the noise-corrupted value of objective function at any chosen point in the parameter space is available. We first introduce a new continuous action-set learning automaton (CALA) and theoretically study its convergence properties, which implies the convergence to the optimal action. Then we give an algorithm, which needs only one function evaluation in each stage, for optimizing an unknown function.