Structured Difference-of-Q via Orthogonal Learning

Defu Cao, Angela Zhou

Offline reinforcement learning is important in many settings with available observational data but the inability to deploy new policies online due to safety, cost, and other concer…