Towards a Theory of Exploration in Continuous-Time Reinforcement Learning

Watch a comprehensive lecture where Professor Harsha Honnappa from Purdue University explores the theoretical framework of continuous-time reinforcement learning in stochastic environments. Delve into the complexities of exploration processes using rough path theory, examining how traditional reinforcement learning methods must be adapted for ultra-high frequency interactions. Learn about the challenges of continuous-time exploration, the application of relaxed controls in Wasserstein space, and the development of pathwise relaxed control frameworks. Discover how these concepts apply to real-world scenarios like stock market volatility modeling and controlled stochastic networks with heavy-tailed service. Understand the mathematical foundations behind establishing value function existence and uniqueness through rough Hamilton-Jacobi-Bellman equations, and explore the implications for entropy-regularized objectives in reinforcement learning.