Current browse context:
cs.LG
Change to browse by:
References & Citations
Computer Science > Machine Learning
Title: Weisfeiler and Leman Go Neural: Higher-order Graph Neural Networks
(Submitted on 4 Oct 2018 (v1), revised 16 Nov 2018 (this version, v2), latest version 30 Nov 2021 (v5))
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. Up to now, GNNs have only been evaluated empirically---showing promising results. The following work investigates GNNs from a theoretical point of view and relates 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.
Submission history
From: Christopher Morris [view email][v1] Thu, 4 Oct 2018 14:31:57 GMT (73kb,D)
[v2] Fri, 16 Nov 2018 12:52:37 GMT (80kb,D)
[v3] Fri, 14 Feb 2020 15:55:24 GMT (317kb,D)
[v4] Tue, 23 Nov 2021 18:51:27 GMT (235kb,D)
[v5] Tue, 30 Nov 2021 15:37:55 GMT (314kb,D)
Link back to: arXiv, form interface, contact.