Carleton University
Technical Report TR-26
May 1983
On the Essential Equivalence of Two Families of Learning Automata
M.A.L. Thathachar & B.J. Oommen
Abstract
Fixed Structure Stochastic Automata (FSSA) have been used to learn the best of a finite set of actions by interacting with a random environment. Two families of such automata are the Tsetlin Automata and the Krylov Automata. In this paper, it is shown that a Krylov Automaton which possesses a certain number of states and which interacts with an environment E1 is equivalent to a Tsetlin Automation possessing the same number of states but which interacts with an environment E2. The relationship between the environments has also been derived. A tremendous gain in computation can thus be obtained in the study of the Krylov Automaton (which is essentially stochastic) by studying the corresponding deterministic Tsetlin automaton in the modified environment. Apart from being of computational significance, this demonstrates a new way of studying certain families of Fixed Structure Stochastic Automata (FSSA) using deterministic automata thus simplifying both the analysis and the computation.