Time: Wednesday 30-Sept-2020 16:00
Live in 11 days & 01:08:41
Motivation / Abstract
Pascal will present two techniques to automatically infer high-level object information from low-level pixel intensities and raw text to accelerate the optimization of a good policy in deep reinforcement learning. The first approach, called MOREL (Motion Oriented REinforcement Learning), consists of a self-supervised technique that learns to detect moving objects in video games. The second approach also consists of a self-supervised technique that automatically infers a belief graph of objects and relations described in text. These techniques allow RL agents to reason at a high level and therefore need less interaction with the environment to find good policies. Furthermore, we can gain insights and some degree of explainability into the resulting policies by inspecting the objects they depend on. The techniques will be demonstrated in Atari Games and TextWorld.