League Training
A training method in multi-agent reinforcement learning in which agents compete against a league composed of diverse strategies and opponents of varying strength. This approach prevents convergence to local optima and encourages the development of general, robust policies. The method was applied at large scale by DeepMind in AlphaStar.
Timeline (last 90 days)
League play has agents compete against a league containing opponents of various strategies and different strengths, in order to prevent them from falling into a local optimum of a single strategy and overfitting.
The concept of league play was applied at scale in DeepMind's StarCraft II AI, AlphaStar.
League play uses population diversity to avoid the rock-paper-scissors problem of cyclic dominance.
With the help of reward shaping and league play, the author ultimately trained agents to exhibit relatively interesting combat behavior.
All Facts (4)
League play has agents compete against a league containing opponents of various strategies and different strengths, in order to prevent them from falling into a local optimum of a single strategy and overfitting.
50%UnverifiedThe concept of league play was applied at scale in DeepMind's StarCraft II AI, AlphaStar.
50%UnverifiedLeague play uses population diversity to avoid the rock-paper-scissors problem of cyclic dominance.
50%UnverifiedWith the help of reward shaping and league play, the author ultimately trained agents to exhibit relatively interesting combat behavior.
50%