This paper considers the problem of using a stochastic automaton to learn the optimal action from a finite set of actions. The automaton learns the action while interacting with an environment. Author proposes a family of automata termed as Generalized Krylov (GK) automata and formalize the learning properties of these automata. The technique for studying GK automata has also been discussed. The advantages of using GK Automata in learning from nonstationary environments have been discussed. The learning and 'unlearning' properties of these automata have been demonstrated.

Additional Metadata
Conference Modeling and Simulation, Volume 15: Proceedings of the Fifteenth Annual Pittsburgh Conference. Part 5: General Modeling and Simulation.
Oommen, J. (1984). APPLICABILITY OF GENERALIZED KRYLOV AUTOMATA TO LEARNING IN NONSTATIONARY ENVIRONMENTS. In Modeling and Simulation, Proceedings of the Annual Pittsburgh Conference (pp. 789–798).