Active learning of timed automata with unobservable resets

TL;DR

This paper introduces a novel active learning approach for timed automata that handles unobservable resets by generalizing existing methods to a new class called reset-free event-recording automata, enabling more effective inference.

Contribution

It extends active learning techniques to general timed automata by developing methods to infer unobservable resets, using the concept of invalidity for efficient hypothesis pruning.

Findings

Proposes reset-free event-recording automata as a new class.

Develops algorithms for on-the-fly detection of invalid reset hypotheses.

Enables active learning of more general timed automata with unobservable resets.

Abstract

Active learning of timed languages is concerned with the inference of timed automata from observed timed words. The agent can query for the membership of words in the target language, or propose a candidate model and verify its equivalence to the target. The major difficulty of this framework is the inference of clock resets, central to the dynamics of timed automata, but not directly observable. Interesting first steps have already been made by restricting to the subclass of event-recording automata, where clock resets are tied to observations. In order to advance towards learning of general timed automata, we generalize this method to a new class, called reset-free event-recording automata, where some transitions may reset no clocks. This offers the same challenges as generic timed automata while keeping the simpler framework of event-recording automata for the sake of readability. Central to our contribution is the notion of invalidity, and the algorithm and data structures to deal with it, allowing on-the-fly detection and pruning of reset hypotheses that contradict observations, a key to any efficient active-learning procedure for generic timed automata.