A Training Algorithm for Systems Described by Stochastic Transition Matrices

Abstract
Stochastic transition matrices are a convenient means for describing the behavior of adaptive and learning systems. Several systems which utilize these matrices and associated reinforcement (reward and punishment) techniques have been reported. A training algorithm is described which has been applied to a learning system described by stochastic transition matrices in which the environment was unknown a priori and nonstationary.

This publication has 4 references indexed in Scilit: