∊-Optimal nonlinear reinforcement scheme under a nonstationary muititeacher environment

Learning behaviours of variable-structure stochastic automata operating in a nonstationary multiteacher environment are considered. As an extended form of the GAE reinforcement scheme, the MGAE scheme is proposed as a reinforcement scheme for a multiteacher environment from which stochastic automata receive responses having arbitrary numbers between 0 and 1. It is shown that the MGAE scheme is ϵ-optimal in the nonstationary multiteacher environment.