Can I read ReBeL: RL and Search for Imperfect Games on EtoBox?

ReBeL: RL and Search for Imperfect Games by jbpkr1 is a document available to read on EtoBox.

What is ReBeL: RL and Search for Imperfect Games about?

The document presents ReBeL, a framework that combines deep reinforcement learning and search to achieve superhuman performance in imperfect-information games like poker. ReBeL expands the concept of "state" to include probabilistic beliefs about the actual state based on common knowledge. It trains value and policy networks on these expanded states through self-play reinforcement learning, and also uses the networks to guide search during training. ReBeL provably converges to a Nash equilibrium in two-play

Author

jbpkr1

Language

EN