Comparison of a stochastic automaton and a related sample mean approach to parameter optimization problems

Stochastic Automata have been proposed as a suitable approach for Adaptive parameter optimization problems with multimodal performance criteria. A recently developed automaton structure [1] with the desired behavioral properties is presented and then contrasted with the most straightforward global strategy, that of Sample Mean estimation. This comparison, which is based on both the cost of sampling and also on the total number of samples, establishes a general point of view within which to assess the advantages of the automaton learning structure approach over the pure sampling approach which in effect, is a non-sequential procedure with no inherent learning capability.