Towards Safe and Efficient Learning in the Physical World - Stanford Seminar

Explore cutting-edge approaches to safe and efficient learning in physical environments through this Stanford seminar featuring Andreas Krause of ETH Zurich. Delve into safe Bayesian optimization, which guarantees both safety and convergence to reachable optima under certain conditions. Examine Bayesian model-based deep reinforcement learning, utilizing epistemic uncertainty in world models for guided exploration while ensuring safety. Discover techniques for meta-learning flexible probabilistic models from related tasks and simulations. See these approaches applied to real-world scenarios, including robotics tasks and tuning the SwissFEL Free Electron Laser. This 46-minute talk, part of Stanford's Robotics and Autonomous Systems Seminar series, offers valuable insights for researchers and practitioners in the field of autonomous systems and machine learning.