Learning Fast with No Goals - VISR Explained

Learning Fast with No Goals - VISR Explained

Explore the concept of rapid generalization in Reinforcement Learning through a 36-minute video lecture on VISR (Fast Task Inference with Variational Intrinsic Successor Features). Delve into successor features, goal-conditioned policies, and their application in adapting to new tasks within the no-reward regime. Examine the weaknesses of successor features, the role of mutual information, and the process of combining successor features. Gain insights into the paper's overview and results, understanding how this DeepMind research contributes to advancing RL capabilities for quick task adaptation.