Who Needs to Know? Minimal Knowledge for Optimal Coordination

07 Jul 2023

To optimally coordinate with others in cooperative games, it is often crucial to have information about one’s collaborators. However, not every feature of collaborators is strategically relevant. In their paper for the International Conference on Machine Learning (ICML) Who Needs to Know? Minimal Knowledge for Optimal Coordination, Niklas Lauffer, Ameesh Shah, Micah Carroll, Michael Dennis and Stuart Russell show that there is a well-defined dichotomy between strategically relevant and irrelevant information and that it can be efficiently computed via a Bellman backup operator in dynamic games. They apply their algorithms to analyze the strategically relevant information for tasks in Overcooked environments. Theoretical and empirical results show that their algorithms are significantly more efficient than baselines. Videos are available here.