Deep Q-Learning for Nash Equilibria: Nash-DQN

Model-free learning for multi-agent stochastic games is an active area of research. Existing reinforcement algorithms, however, are often restricted to zero-sum games, and applicable only in small state-action spaces or other simplified settings. Here, we develop a new data efficient Deep-Q-learning methodology model-free Nash equilibria general-sum games. The algorithm uses local linear-quadratic expansion the game, which leads analytically solvable optimal actions. parametrized by deep neural networks give it sufficient flexibility learn environment without need experience all pairs. We study symmetry properties stemming from label-invariant as proof concept, apply our trading strategies competitive electronic markets.