Training Agents using Cheap Data Sources

I gave an invited talk at Princeton Reinforcement Learning Lab. I talk about how to leverage cheap action-free data to improve the generalization of reinforcement learning policies. Here are the slides.