My research has focused on studying the dynamics behind noisy systems and their applications to machine learning, specifically in game theory. I have been extending past results on asynchronous stochastic approximations in order to build a framework to study learning algorithms in stochastic games. The aim is to provide an actor-critic algorithm which converges to a Nash equilibrium for N-player identical interest and 2-player zero-sum, discounted reward stochastic games. I have also been using the abstract stochastic approximation framework to study stochastic fictitious play in games with continuous action sets.
Office 3.3, School of Mathematics,
University of Bristol,
University Walk,
Clifton,
Bristol BS8 1TW
Tel: 0117 3318275
Email: Steven.Perkins.09@bristol.ac.uk