L*-Based Learning of Markov Decision Processes (Extended Version)

TL;DR

This paper extends L*-based automata learning techniques to deterministic Markov decision processes, proposing a novel sampling-based algorithm that learns complete model structures and outperforms passive methods in accuracy.

Contribution

It introduces a new L*-based learning algorithm for Markov decision processes that relaxes perfect information assumptions and learns full model structures from sampled traces.

Findings

Sampling-based algorithm achieves higher accuracy than passive methods.

The algorithm learns complete model structures including states.

Experiments validate improved performance with the same test data.

Abstract

Automata learning techniques automatically generate system models from test observations. These techniques usually fall into two categories: passive and active. Passive learning uses a predetermined data set, e.g., system logs. In contrast, active learning actively queries the system under learning, which is considered more efficient. An influential active learning technique is Angluin's L* algorithm for regular languages which inspired several generalisations from DFAs to other automata-based modelling formalisms. In this work, we study L*-based learning of deterministic Markov decision processes, first assuming an ideal setting with perfect information. Then, we relax this assumption and present a novel learning algorithm that collects information by sampling system traces via testing. Experiments with the implementation of our sampling-based algorithm suggest that it achieves better accuracy than state-of-the-art passive learning techniques with the same amount of test data. Unlike existing learning algorithms with predefined states, our algorithm learns the complete model structure including the states.