Asynchronous Advantage Actor-Critic

A3C runs multiple agents in parallel, each interacting with its own copy of the environment. This parallelization provides diverse experience and stable learning. It uses an advantage function to reduce variance in policy gradient estimates and combines this with asynchronous updates to a global network.

Distributed training systems, game AI, robot control, parallel simulations