A Nonlinear Reinforcement Scheme for Stochastic Learning Automata

A stochastic automaton can perform a finite number of actions in a random environment. When a specific action is performed, the environment responds by producing an environment output that is stochastically related to the action. This response may be favorable or unfavorable. The aim is to design an automaton that can determine the best action guided by past actions and responses. The reinforcement scheme presented is shown to satisfy all necessary and sufficient conditions for absolute expediency for a stationary environment. An automaton using this scheme is guaranteed to „do better” at every time step than at the previous step (expected value of the average penalty at one iteration step is less than of the previous step for all steps). Some simulation results are presented, which prove that our algorithm converges to a solution faster than the one given in [7]. Key-Words: Stochastic Learning Automata, Reinforcement Learning