Learning of Timed Systems

The goal of this project is to extend the learning algorithm of Angluin and others to the setting of timed systems. We study event-recording automata (ERAs). These are timed automata that, for every action a, use a clock that records the time of the last occurence of a. Event-recording automata are sufficiently expressive to model many interesting timed systems; for instance, they are as powerful as timed transition systems. One of our approaches is based on the idea to reuse the techniques of learning regular systems instead of learning timed systems directly. This approach requires region graph construction. We are also interested in the problem of learning a smaller systems. To achieve this goal we unify the queried information when it is "similar" which results in merging states in the automaton construction.