Learning Mealy Machines with Timers

Bengt Jonsson and Frits Vaandrager. Learning Mealy Machines with Timers January 2018.

Abstract

We introduce a new model of Mealy machines with timers (MMTs), which is able to describe the timing behavior of a broad class of practical systems, and sufficiently restricted for active learning algorithms. We present a natural extension of Angluin's active learning algorithm, which employs sequences of inputs with precise timing. Our algorithm is based on three key results: (i) an untimed semantics for MMTs, which is equivalent to the natural timed one (ii) a Nerode congruence based on the untimed semantics, and (iii) an active automata learning algorithm which is based on approximating this Nerode congruence. This algorithm allows to learn MMTs using a number of membership and equivalence queries, which is polynomial in the number of states of the resulting MMT, and doubly exponential in the maximal number of simultaneously active timers.

Paper (pdf)
Slides (pdf)