Learning Automata and Their Applications to Intelligent Systems

Stochastic ranking and selection aim to design statistical procedures that select a candidate with the highest mean performance from a finite set of candidates whose performance is uncertain but may be estimated by a learning process based on interactions with a stochastic environment or by simulations. The number of interactions with environments or simulations of candidates is usually limited and needs to be minimized due to limited computational resources. Traditional approaches taken in the literature include frequentist statistics, Bayesian statistics, heuristics, and asymptotic convergence in probability. Novel and recent approaches to stochastic ranking and selection problems are learning automata and ordinal optimization.