Hub seminar series
Xinyu Li, a new Postdoctoral Research Associate based at Oxford’s Mathematical Institute, delivered the latest in the hub’s seminar series on 23 October. The seminars provide a great opportunity for the newest members of the hub’s teams to present their research to the community.
Xinyu’s talk, entitled Markov α-Potential Games: A Framework to study Multi-Agent Reinforcement Learning, proposed a new framework of Markov α-potential games to study Markov games. It showed that any Markov game with finite-state and finite-action is a Markov α-potential game, and established the existence of an associated α-potential function. Any optimizer of an α-potential function is shown to be an α-stationary Nash equilibrium.
Xinyu studied two important classes of practically significant Markov games, Markov congestion games and the perturbed Markov team games, via the framework of Markov α-potential games, with explicit characterisation of an upper bound for α and its relation to game parameters. She also provided a semi-infinite linear programming-based formulation to obtain an upper bound for α for any Markov game. Furthermore, Xinyu studied two equilibrium approximation algorithms, namely the projected gradient- ascent algorithm and the sequential maximum improvement algorithm, along with their Nash regret analysis.