Finite-Time Last-Iterate Convergence for Learning in Multi-Player Games

Neural Information Processing Systems 

The standard metric to quantify an online learning algorithm's performance is the regret.