On Parallel Induction of Nondeterministic Finite Automata

The paper discusses the problem of the induction of a minimal nondeterministic finite automaton (INFA) consistent with a given set of examples and counterexamples. The main contribution of the paper is the proposal of an efficient parallel algorithm transforming the NIA induction problem into a family of constraint satisfaction problems (CSP). Two original techniques for fast CSPs evaluation are proposed and discussed. The efficacy of the parallel algorithm is evaluated experimentally on selected benchmarks. The proposed algorithm solves all analyzed benchmark examples, so called Tomita languages, in the time below half a minute, which should be considered an important achievement, as compared to our previous efforts which required minutes or hours to solve some of the aforementioned benchmarks.