An automaton decomposition for learning system environments

A finite automaton two-component cascade decomposition is presented in which the first component has a synchronizer and the second component is a permutation automaton. The synchronizer corresponds to a primitive idempotent element e in the transition monoid M of the automaton. The state set of the second component is the range of e; each state of the first component is an image of this range under one of the transitions in M. The transition monoid of the second component is the group eMe. as a conceptual tool, the decomposition can be used to clarify the credit assignment problems faced by learning system reward schemes in finite automaton environments.

Authors

• T.H. Westerdale

Keywords

No keywords are indexed for this paper.