General semiring conditions for weighted L* termination

Abstract

Weighted automata learning has been well-studied over (semi)rings which have a dimension theory. In this work, we investigate active learning for weighted automata over semirings that lack such a structure. We first show that the principal ideal domain variant of the L* algorithm also terminates for a larger class of semirings containing Noetherian semirings, commutative rings, and bounded distributed lattices. We then present a variant of this algorithm for learning deterministic weighted automata, and prove that this algorithm terminates for commutative semirings.