Article Source

• Graph Neural Networks and Applications to Deep Reinforcement Learning

Graph Neural Networks and Applications to Deep Reinforcement Learning

Abstract

Many interesting problems involve processing information that’s best represented by graphs. To that end, prior work has introduced Graph Neural Networks (GNNs) that bring the powerful capabilities of deep learning to graphical inputs. In this talk, I will provide an overview of various GNN techniques, including GCNs, as well as applications seen in recent research in Deep Reinforcement Learning (RL). I will go in depth on using graphs to represent common-sense, prior knowledge about the world, and how that can be leveraged for faster learning.