Optimal last-iterate convergence in matrix games with bandit feedback using the log-barrier
The research investigates achieving last-iterate convergence for minimax policies in zero-sum matrix games under online, bandit feedback. By combining mirror descent with log-barrier regularization and a dual-focused analysis, the method establishes an optimal convergence rate of O(˜t−1/4) with high probability in the bandit setting, extending this result to extensive-form games as well.