A Properties and separations for generalized equilibria A.1 Proof of Proposition 1 Proof. The set of (Φ

If a player receives substantially more or less than the corresponding value, this would imply a violation of the regret constraints for at least one of the players' learning algorithms. Theorem 8. F or each of the following, there exists a The corresponding values for this game are simple to compute: 1. Val If either player violates the other's trust o (T) times, then the player defects to playing L First we elaborate upon the trusting phase. This process repeats for every window. We now show that this algorithm satisfies both conditions in the theorem statement. The last step follows since NumWindows(T) o (T), because Length (T) o (T).