Exploration and exploitation trade-off in multiagent learning

This paper focuses on the trade-off between exploration and exploitation in multiagent learning and explores some fundamental factors that contribute to clarifying this trade-off. Through inventive simulations on distributed constraint satisfaction problems in multiagent environments, the following implications are revealed: (1) the trade-off between exploration and exploitation at the collective level is not easy to be solved when considering the trade-off at the individual level; but (2) the trade-off at the collective level can be solved by introducing a simple gradient search in the trade-off at the individual level.