GA and ILS for Optimizing the Size of NFA Models

Grammatical inference consists in learning a formal grammar (as set of rewrite rules or finite state machine). We are concerned with Nondeterministic Finite Automata (NFA) given size from samples positive and negative words. NFA can naturally be modeled SAT. The standard model [1] being enormous, we also try based on prefixes [2] which generates smaller instances. propose new suffixes hybrid suffixes. then focus optimizing the generated SAT instances issued models. present two techniques to optimize this combination, one Iterated Local Search (ILS), second Genetic Algorithm (GA). Optimizing combination significantly reduces their solving time, but at cost longer generation time. We, therefore, study balance between time thanks some experimental comparisons, analyze our various improvements.