Titre: Inductive Reasoning and Representation Learning on Knowledge Graphs

Conférencier: William L. Hamilton

Résumé: New developments in techniques for knowledge graph embedding and graph neural networks have pushed the state-of-the-art in wide variety of applications, including knowledge base completion, natural language question answering, and recommendation engines. However, most existing approaches require that all entities in the knowledge graph are known in advance; these previous approaches are inherently transductive and do not naturally generalize to unseen (sub)graphs. In this talk, I will discuss the challenge of inductive representation learning and reasoning on knowledge graphs, where the goal is to develop machine learning methods that can generalize to previously unseen (sub)graphs after training. I will discuss initial attempts to use graph neural networks for inductive learning on knowledge graphs, a new benchmark dataset that explicitly tests inductive reasoning capabilities, and I will highlight connections between this work and the field of inductive logic programming.

Bio: William L. Hamilton (wlh@cs.mcgill.ca) is an Assistant Professor in McGill University's School of Computer Science and a Core Professor at the Mila AI Institute of Quebec. He completed his PhD at Stanford University in 2018, and his research focuses on graph representation learning, as well as applications in large-scale computational social science and biology. He was the SAP Stanford Graduate Fellow from 2014 to 2018, received the Cozzarelli Best Paper Award from the Proceedings of the National Academy of Sciences (PNAS) in 2017, and his work has been featured in numerous media outlets, including Wired, The New York Times, and The BBC.

Bienvenue à tous!