Bridging State and Action: Towards Continual Reinforcement Learning - 2023 Fall Robotics Colloquium

Explore a cutting-edge lecture on continual reinforcement learning in this 50-minute robotics colloquium featuring Khimya Khetarpal from Google Deepmind. Delve into the development of AI systems that efficiently represent world knowledge, plan, and adapt to changes over time through learning and interaction. Discover a framework enabling AI agents to represent and reason about their environment using affordances, which reduce action possibilities and facilitate efficient transition model learning. Learn about an approach for building temporally abstract partial models using affordances, and understand the trade-offs between single-step and temporally extended partial models in planning and learning. Gain insights into the benefits of affordances, including faster planning across timescales, improved sample efficiency for learning transition models, and robust generalization. Presented by an accomplished researcher recognized as a Rising Star in EECS and awarded for her work in lifelong learning, this talk offers valuable perspectives on advancing artificial intelligence and reinforcement learning.