 
Summary: Dynamic Programming with Stochastic Opponent Models in Social Games
TszChiu Au
Department of Computer Science
University of Maryland
College Park, MD 20742, U.S.A.
chiu@cs.umd.edu
Abstract
Policy makers often confront with the following problem:
how best their organization can repeatedly interact with other
organizations such that the longterm utility of their organi
zation can be maximized? This problem is difficult because
policy makers usually know very little about other organi
zations, and therefore they cannot make perfect predictions
about the other organizations' behaviors. In this paper, we
formulate this problem as social games in which (1) there are
two or more agents interacting with each other; (2) each agent
can perform more than one action in each interaction; and (3)
the payoff matrix is not fixed; the payoff matrix varies from
one situation to another. We devised a dynamic program
ming algorithm to compute a policy given the model of the
