WL meet VC

Christopher Morris, Floris Geerts, Jan Tönshoff, Martin Grohe
Proceedings of the 40th International Conference on Machine Learning, PMLR 202:25275-25302, 2023.

Abstract

Recently, many works studied the expressive power of graph neural networks (GNNs) by linking it to the $1$-dimensional Weisfeiler-Leman algorithm ($1\text{-}\mathsf{WL}$). Here, the $1\text{-}\mathsf{WL}$ is a well-studied heuristic for the graph isomorphism problem, which iteratively colors or partitions a graph’s vertex set. While this connection has led to significant advances in understanding and enhancing GNNs’ expressive power, it does not provide insights into their generalization performance, i.e., their ability to make meaningful predictions beyond the training set. In this paper, we study GNNs’ generalization ability through the lens of Vapnik-Chervonenkis (VC) dimension theory in two settings, focusing on graph-level predictions. First, when no upper bound on the graphs’ order is known, we show that the bitlength of GNNs’ weights tightly bounds their VC dimension. Further, we derive an upper bound for GNNs’ VC dimension using the number of colors produced by the $1\text{-}\mathsf{WL}$. Secondly, when an upper bound on the graphs’ order is known, we show a tight connection between the number of graphs distinguishable by the $1\text{-}\mathsf{WL}$ and GNNs’ VC dimension. Our empirical study confirms the validity of our theoretical findings.

Cite this Paper

BibTeX
@InProceedings{pmlr-v202-morris23a, title = {{WL} meet {VC}}, author = {Morris, Christopher and Geerts, Floris and T\"{o}nshoff, Jan and Grohe, Martin}, booktitle = {Proceedings of the 40th International Conference on Machine Learning}, pages = {25275--25302}, year = {2023}, editor = {Krause, Andreas and Brunskill, Emma and Cho, Kyunghyun and Engelhardt, Barbara and Sabato, Sivan and Scarlett, Jonathan}, volume = {202}, series = {Proceedings of Machine Learning Research}, month = {23--29 Jul}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v202/morris23a/morris23a.pdf}, url = {https://proceedings.mlr.press/v202/morris23a.html}, abstract = {Recently, many works studied the expressive power of graph neural networks (GNNs) by linking it to the $1$-dimensional Weisfeiler-Leman algorithm ($1\text{-}\mathsf{WL}$). Here, the $1\text{-}\mathsf{WL}$ is a well-studied heuristic for the graph isomorphism problem, which iteratively colors or partitions a graph’s vertex set. While this connection has led to significant advances in understanding and enhancing GNNs’ expressive power, it does not provide insights into their generalization performance, i.e., their ability to make meaningful predictions beyond the training set. In this paper, we study GNNs’ generalization ability through the lens of Vapnik-Chervonenkis (VC) dimension theory in two settings, focusing on graph-level predictions. First, when no upper bound on the graphs’ order is known, we show that the bitlength of GNNs’ weights tightly bounds their VC dimension. Further, we derive an upper bound for GNNs’ VC dimension using the number of colors produced by the $1\text{-}\mathsf{WL}$. Secondly, when an upper bound on the graphs’ order is known, we show a tight connection between the number of graphs distinguishable by the $1\text{-}\mathsf{WL}$ and GNNs’ VC dimension. Our empirical study confirms the validity of our theoretical findings.} }
Endnote
%0 Conference Paper %T WL meet VC %A Christopher Morris %A Floris Geerts %A Jan Tönshoff %A Martin Grohe %B Proceedings of the 40th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2023 %E Andreas Krause %E Emma Brunskill %E Kyunghyun Cho %E Barbara Engelhardt %E Sivan Sabato %E Jonathan Scarlett %F pmlr-v202-morris23a %I PMLR %P 25275--25302 %U https://proceedings.mlr.press/v202/morris23a.html %V 202 %X Recently, many works studied the expressive power of graph neural networks (GNNs) by linking it to the $1$-dimensional Weisfeiler-Leman algorithm ($1\text{-}\mathsf{WL}$). Here, the $1\text{-}\mathsf{WL}$ is a well-studied heuristic for the graph isomorphism problem, which iteratively colors or partitions a graph’s vertex set. While this connection has led to significant advances in understanding and enhancing GNNs’ expressive power, it does not provide insights into their generalization performance, i.e., their ability to make meaningful predictions beyond the training set. In this paper, we study GNNs’ generalization ability through the lens of Vapnik-Chervonenkis (VC) dimension theory in two settings, focusing on graph-level predictions. First, when no upper bound on the graphs’ order is known, we show that the bitlength of GNNs’ weights tightly bounds their VC dimension. Further, we derive an upper bound for GNNs’ VC dimension using the number of colors produced by the $1\text{-}\mathsf{WL}$. Secondly, when an upper bound on the graphs’ order is known, we show a tight connection between the number of graphs distinguishable by the $1\text{-}\mathsf{WL}$ and GNNs’ VC dimension. Our empirical study confirms the validity of our theoretical findings.
APA
Morris, C., Geerts, F., Tönshoff, J. & Grohe, M.. (2023). WL meet VC. Proceedings of the 40th International Conference on Machine Learning, in Proceedings of Machine Learning Research 202:25275-25302 Available from https://proceedings.mlr.press/v202/morris23a.html.

Related Material