L*-Based Learning of Probabilistic Timed Automata with One Clock (S)

Abstract

Probabilistic timed automata (PTAs) allow for analyzing systems which operate under the existence of both probability and time. Probabilistic model checking for PTAs requires formal models as a basis. In this paper, we consider a learning algorithm for PTAs with one clock (OPTAs) which extends Angluin's L * algorithm. For the black-box systems, we estimate the probabilities by sampling system traces, and apply the framework of Probably Approximately Correct (PAC) learning to provide correctness guarantees respect to error and confidence parameters. Experiments with the implementation of our learning algorithm include the sender of CSMA/CD protocol and randomly generated examples. Evaluating results show the effectiveness of our learning algorithm.