In this presentation, Dr. Jonathan Dodge covers Neurosymbolic Learning with Formally Verified Exploration, by Anderson et al. The paper introduces the concept of formally verified exploration in neurosymbolic reinforcement learning. The main goal is to ensure a safe policy for exploring environments, e.g., to prevent robots from falling into pits before they know that they should avoid falling into pits and thus damaging the robot before it can explore sufficiently.
Presentation Link: https://psu.mediaspace.kaltura.com/media/Group+Meeting/1_xixwp4ql