The multiagent learning research group at Brown studies automated agents that engage in game-theoretic (i.e., strategic) activities, exemplified by, for example, the Prisoners' Dilemma. While participating in such games, agents interact repeatedly with their environment and with one another, giving them an opportunity to employ machine learning techniques to learn to make decisions that best satisfy their objectives.

Research on multiagent learning in games at Brown encompasses the following:

• designing learning algorithms that behave optimally in the non-stationary environments that naturally arise in multiagent settings,
• envisioning new game-theoretic solution concepts that capture the collective behavior of adaptive agents that employ heuristic learning strategies, and
• creating market mechanisms that induce collective agent behavior which satisfies globally desirable properties, such as fairness and efficiency.