Probabilistic Search as a Strategy Selection Procedure

An alternative solution to the problem of the selection of the best strategy in a random environment is presented by using a probabilistic search procedure. The asymptotic optimality of the technique is proved, and a brief comparison with stochastic automata with variable structures is made. A specific organization of the optimal search procedure is developed based on continued learning of some statistics of the random environment, and it is shown to be fast-converging, powerful in high noise random environments, and insensitive to search parameter selection.