A PAC Learning Algorithm for LTL and Omega-regular Objectives in MDPs
Machine Learning
2024-02-22 v3 Logic in Computer Science
Abstract
Linear temporal logic (LTL) and omega-regular objectives -- a superset of LTL -- have seen recent use as a way to express non-Markovian objectives in reinforcement learning. We introduce a model-based probably approximately correct (PAC) learning algorithm for omega-regular objectives in Markov decision processes (MDPs). As part of the development of our algorithm, we introduce the epsilon-recurrence time: a measure of the speed at which a policy converges to the satisfaction of the omega-regular objective in the limit. We prove that our algorithm only requires a polynomial number of samples in the relevant parameters, and perform experiments which confirm our theory.
Cite
@article{arxiv.2310.12248,
title = {A PAC Learning Algorithm for LTL and Omega-regular Objectives in MDPs},
author = {Mateo Perez and Fabio Somenzi and Ashutosh Trivedi},
journal= {arXiv preprint arXiv:2310.12248},
year = {2024}
}