Hierarchical Reinforcement Learning
Doina discusses the advantages of hierarchical reinforcement learning in multitask environments, emphasizing the importance of slow initial learning for efficient exploration in future tasks. She draws an analogy between hierarchical RL and CNNs, highlighting how both operate at different levels of resolution—one in actions and the other in features—while acknowledging the variability in time scales that complicates this comparison. The conversation illustrates the need for adaptability in decision-making, akin to managing both quick reactions and longer processes, like cooking.In this clip
From this podcast
The TWIML AI Podcast (formerly This Week in Machine Learning & Artificial Intelligence)
Hierarchical and Continual RL with Doina Precup - #567
Related Questions