Learning to Explore with Scalable Supervision - IPAM at UCLA

Explore the challenges and solutions in reinforcement learning for robotic control tasks in this 25-minute conference talk by Lisa Lee from Google Brain. Delve into efficient exploration techniques in high-dimensional spaces, learning efficiency, safety concerns, and reducing the cost of human supervision. Discover how to balance self-supervised and human-supervised reinforcement learning to train agents effectively. Examine methods for amortizing exploration costs, developing semantically meaningful representations for faster learning, and distilling exploration into reusable reward functions. Gain insights from this presentation at IPAM's Mathematics of Collective Intelligence Workshop, recorded on February 16, 2022, at the Institute for Pure & Applied Mathematics (IPAM) at UCLA.