AI & Fundamentals

Graph Neural Networks and Graph Isomorphism - Will Hamilton, Assistant Professor, McGill University

Abstract:

In recent years, graph neural networks (GNNs) have emerged as a powerful neural architecture to learn vector representations of nodes and graphs in a supervised, end-to-end fashion. However, GNNs have mainly been evaluated empirically---showing promising results. This talk will discuss GNNs from a theoretical point of view and relate them to the 1-dimensional Weisfeiler-Leman graph isomorphism heuristic (1-WL). We show that GNNs have the same expressiveness as the 1-WL in terms of distinguishing non-isomorphic (sub-)graphs. Hence, both algorithms also have the same shortcomings. Based on this, we propose a generalization of GNNs, so-called k-dimensional GNNs (k-GNNs), which can take higher-order graph structures at multiple scales into account. These higher-order structures play an essential role in the characterization of social networks and molecule graphs. Our experimental evaluation confirms our theoretical findings as well as confirms that higher-order information is useful in the task of graph classification and regression.

Bio

William L. Hamilton is an Assistant Professor in the School of Computer Science at McGill University, Canada CIFAR Chair in AI, and a member of the Mila - Quebec AI Institute. He received the 2018 Arthur Samuel Thesis Award for the top Computer Science PhD Thesis from Stanford University, as well as the 2014 CAIAC MSc Thesis Award for best AI-themed MSc thesis in Canada and a Best Paper Award from the Proceedings of the National Academy of Sciences (PNAS) in 2017. His primary research interests relate to the quickly growing machine learning subfields of graph representation learning and graph neural networks. 

Host:

Professor David Poole, Computer Science, UBC

CAIDA Contact: 

Arynn Keane

arynnk@mail.ubc.ca