Scaling up Mean Field Games with Online Mirror Descent


We will explore numerical methods for large mean field games and their convergence to possibly weak equilibrium. We will improve the state of the art over fictitious play and best response algorithms. Our approach relies on mirror descent type algorithms and performs well in high dimensional mean field games.

theoretical convergence to Nash equilibrium convergence to coarse correlated equilibria under weaker assumptions Numerical algorithms in several environments for classical mean field game models in high dimension