Learning Markov Decision Processes Based on Genetic Programming

Abstract

Model checking is used to verify the security of communication protocols in which the behavior is stochastic influenced by the environment. Automata learning settles the problem of obtaining formal models from observable data of black-box systems. It is available for different variations of finite automata to in model checking. Genetic Programming is a machine learning technique that automatically generates programs and outputs a fittest program. In this paper, we present an approach to learn markov decision progresses based on the framework of genetic programming. The approach outputs the fittest model with a set of system traces by refining iteratively models. We evaluate our method on one probabilistic system from the literature and 30 randomly generated examples.