Stanford's Shuran Song details how generative control models like diffusion can actively explore for sample-efficient reinforcement learning

Exploration has always been an important part of any RL algorithms. But in the era of generative control (i.e., where the base policy is a diffusion or flow model), should we formulate it differently?

It turns out that generative policies introduce both new opportunities and challenges for exploration. Check out @calvinyluo 🧵👇 below to learn more!