Learning Control from Minimal Prior Knowledge

Explore the cutting-edge research on neural reinforcement learning methods in this 54-minute lecture by Martin Riedmiller from DeepMind Technologies. Delve into the challenges of developing autonomous learning systems for real-world control scenarios, focusing on data-efficient and robust methods. Examine two crucial research areas: highly efficient off-policy learning and effective exploration. Discover examples of learning agent designs capable of mastering increasingly complex tasks from scratch in both simulated and real environments. Gain insights into the future of intelligent systems and their ability to learn control with minimal prior knowledge.

Martin Riedmiller: "Learning Control from Minimal Prior Knowledge"