Reasoning about Strategic Abilities in Stochastic Multi-agent Systems

作     者：Zhang, Yedi Song, Fu Chen, Taolue Wu, Xuzhi 

作者机构：Department of Computing Science National University of Singapore Singapore State Key Laboratory of Computer Science Institute of Software Chinese Academy of Sciences China School of Computing and Mathematical Sciences Birkbeck University of London United Kingdom College of Computer Science and Software Engineering Shenzhen University China 

出 版 物：《arXiv》 (arXiv)

年 卷 期：2024年

核心收录：

主　　题：Stochastic systems 

摘      要：Reasoning about strategic abilities is key to AI systems comprising multiple agents, which provide a unified framework for formalizing various problems in game theory, social choice theory, etc. In this work, we propose a probabilistic extension of the alternating-time μ-calculus (AMC), named PAMC, for reasoning about the strategic abilities of agents in stochastic multi-agent systems. We show that PAMC subsumes two existing logics AMC and PμTL (a probabilistic extension of the modal μ-calculus), but is incomparable with the probabilistic alternating-time temporal logic (PATL). We study the problems of model checking and satisfiability checking for PAMC. We first give a model checking algorithm by leveraging algorithms for solving normal-form games and AMC model checking. We establish that the model checking problem of PAMC remains in UP∩co-UP, the same complexity class as the model checking problem for AMC and PμTL. We also provide a new reduction from the satisfiability problem of PAMC to solving parity games, by which we obtain an EXPTIME decision procedure, as well as the small model property which allows to construct a model for each satisfiable PAMC formula. Satisfiability in PAMC has the same complexity as in the modal μ-calculus, unlike PCTL and PATL whose satisfiability checking problems remain open. We have implemented both the model checking and satisfiability checking algorithms as open-source tools. Experimental results are reported, showcasing the practical applications and effectiveness of our approaches. © 2024, CC BY.