Stochastic game

About: Stochastic game is a research topic. Over the lifetime, 9493 publications have been published within this topic receiving 202664 citations.

An Analysis of Stochastic Game Theory for Multiagent Reinforcement Learning

Abstract: : Learning behaviors in a multiagent environment are crucial for developing and adapting multiagent systems. Reinforcement learning techniques have addressed this problem for a single agent acting in a stationary environment, which is modeled as a Markov decision process (MDP). But, multiagent environments are inherently non-stationary since the other agents are free to change their behavior as they also learn and adapt. Stochastic games, first studied in the game theory community, are a natural extension of MDPs to include multiple agents. In this paper we contribute a comprehensive presentation of the relevant techniques for solving stochastic games from both the game theory community and reinforcement learning communities. We examine the assumptions and limitations of these algorithms, and identify similarities between these algorithms, single agent reinforcement learners, and basic game theory techniques.

An evolutionary approach to pre-play communication

Abstract: We add a round of pre-play communication to a finite two-player game played by a population of players. Pre-play communication is cheap talk in the sense that it does not directly enter the payoffs. The paper characterizes the set of strategies that are stable with respect to a stochastic dynamic adaptive process. Periodically players have an opportunity to change their strategy with a strategy that is more successful against the current population. Any strategy that weakly improves upon the current poorest performer in the population enters with positive probability. When there is no conflict of interest between the players, only the efficient outcome is stable with respect to these dynamics. For general games the set of stable payoffs is typically large. Every efficient payoff recurs infinitely often.

PRISM-games: a model checker for stochastic multi-player games

Abstract: We present PRISM-games, a model checker for stochastic multi-player games, which supports modelling, automated verification and strategy synthesis for probabilistic systems with competitive or cooperative behaviour. Models are described in a probabilistic extension of the Reactive Modules language and properties are expressed using rPATL, which extends the well-known logic ATL with operators to reason about probabilities, various reward-based measures, quantitative properties and precise bounds. The tool is based on the probabilistic model checker PRISM, benefiting from its existing user interface and simulator, whilst adding novel model checking algorithms for stochastic games, as well as functionality to synthesise optimal player strategies, explore or export them, and verify other properties under the specified strategy.