Compositional Automata Learning of Synchronous Systems

Abstract

AbstractAutomata learning is a technique to infer an automaton model of a black-box system via queries to the system. In recent years it has found widespread use both in industry and academia, as it enables formal verification when no model is available or it is too complex to create one manually. In this paper we consider the problem of learning the individual components of a black-box synchronous system, assuming we can only query the whole system. We introduce a compositional learning approach in which several learners cooperate, each aiming to learn one of the components. Our experiments show that, in many cases, our approach requires significantly fewer queries than a widely-used non-compositional algorithm such as $$\mathtt {L^*}$$ L ∗ .