Checking Liveness Properties of Concurrent Systems by Reinforcement Learning

Abstract

In this paper, we propose a new method of testing concurrent systems by using an artificial intelligence technique: reinforcement learning. In particular, the method verifies the liveness properties given in temporal logic formulas and dynamically controls a target system in runtime monitoring to efficiently reveal possible error against specification. In this control, the learning method accumulates necessary information by monitoring the running system. We built a simulator to evaluate this idea and conducted experiments with simple examples. As a result we showed the effectiveness of this approach for solving the difficult problem of testing the liveness properties of concurrent systems.