Beyond Weisfeiler-Lehman: A Quantitative Framework for GNN Expressiveness

Designing expressive Graph Neural Networks (GNNs) is a fundamental topic in the graph learning community. So far, GNN expressiveness has been primarily assessed via the Weisfeiler-Lehman (WL) hierarchy. However, such an expressivity measure has notable limitations: it is inherently coarse, qualitati...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Zhang, Bohang, Gai, Jingchu, Du, Yiheng, Ye, Qiwei, He, Di, Wang, Liwei
Format: Artikel
Sprache:eng
Schlagworte:
Online-Zugang:Volltext bestellen
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
container_end_page
container_issue
container_start_page
container_title
container_volume
creator Zhang, Bohang
Gai, Jingchu
Du, Yiheng
Ye, Qiwei
He, Di
Wang, Liwei
description Designing expressive Graph Neural Networks (GNNs) is a fundamental topic in the graph learning community. So far, GNN expressiveness has been primarily assessed via the Weisfeiler-Lehman (WL) hierarchy. However, such an expressivity measure has notable limitations: it is inherently coarse, qualitative, and may not well reflect practical requirements (e.g., the ability to encode substructures). In this paper, we introduce a unified framework for quantitatively studying the expressiveness of GNN architectures, addressing all the above limitations. Specifically, we identify a fundamental expressivity measure termed homomorphism expressivity, which quantifies the ability of GNN models to count graphs under homomorphism. Homomorphism expressivity offers a complete and practical assessment tool: the completeness enables direct expressivity comparisons between GNN models, while the practicality allows for understanding concrete GNN abilities such as subgraph counting. By examining four classes of prominent GNNs as case studies, we derive simple, unified, and elegant descriptions of their homomorphism expressivity for both invariant and equivariant settings. Our results provide novel insights into a series of previous work, unify the landscape of different subareas in the community, and settle several open questions. Empirically, extensive experiments on both synthetic and real-world tasks verify our theory, showing that the practical performance of GNN models aligns well with the proposed metric.
doi_str_mv 10.48550/arxiv.2401.08514
format Article
fullrecord <record><control><sourceid>arxiv_GOX</sourceid><recordid>TN_cdi_arxiv_primary_2401_08514</recordid><sourceformat>XML</sourceformat><sourcesystem>PC</sourcesystem><sourcerecordid>2401_08514</sourcerecordid><originalsourceid>FETCH-LOGICAL-a674-e8000fa558974788f30e0d7d1ac502ff9ebbef379cef0f712f3f4952adf802263</originalsourceid><addsrcrecordid>eNotz81OwkAUhuHZuDDoBbhybqD1zB8zdYcEkKTBmJC4bA70nDiRtmRaEe5eBFfv4ku-5BHiQUFug3PwhOkYD7m2oHIITtlbsXyhU9fW8oNizxR3lLKSPhtsn-VEvn9jO8QBh3ggOU_Y0E-XviR3SS5WKzk77hP1_Xlsz7kTN4y7nu7_OxLr-Ww9fc3Kt8VyOikzHHubUQAARudC4a0PgQ0Q1L5WuHWgmQvabIiNL7bEwF5pNmwLp7HmAFqPzUg8Xm8vlmqfYoPpVP2ZqovJ_ALT6EcP</addsrcrecordid><sourcetype>Open Access Repository</sourcetype><iscdi>true</iscdi><recordtype>article</recordtype></control><display><type>article</type><title>Beyond Weisfeiler-Lehman: A Quantitative Framework for GNN Expressiveness</title><source>arXiv.org</source><creator>Zhang, Bohang ; Gai, Jingchu ; Du, Yiheng ; Ye, Qiwei ; He, Di ; Wang, Liwei</creator><creatorcontrib>Zhang, Bohang ; Gai, Jingchu ; Du, Yiheng ; Ye, Qiwei ; He, Di ; Wang, Liwei</creatorcontrib><description>Designing expressive Graph Neural Networks (GNNs) is a fundamental topic in the graph learning community. So far, GNN expressiveness has been primarily assessed via the Weisfeiler-Lehman (WL) hierarchy. However, such an expressivity measure has notable limitations: it is inherently coarse, qualitative, and may not well reflect practical requirements (e.g., the ability to encode substructures). In this paper, we introduce a unified framework for quantitatively studying the expressiveness of GNN architectures, addressing all the above limitations. Specifically, we identify a fundamental expressivity measure termed homomorphism expressivity, which quantifies the ability of GNN models to count graphs under homomorphism. Homomorphism expressivity offers a complete and practical assessment tool: the completeness enables direct expressivity comparisons between GNN models, while the practicality allows for understanding concrete GNN abilities such as subgraph counting. By examining four classes of prominent GNNs as case studies, we derive simple, unified, and elegant descriptions of their homomorphism expressivity for both invariant and equivariant settings. Our results provide novel insights into a series of previous work, unify the landscape of different subareas in the community, and settle several open questions. Empirically, extensive experiments on both synthetic and real-world tasks verify our theory, showing that the practical performance of GNN models aligns well with the proposed metric.</description><identifier>DOI: 10.48550/arxiv.2401.08514</identifier><language>eng</language><subject>Computer Science - Data Structures and Algorithms ; Computer Science - Discrete Mathematics ; Computer Science - Learning ; Mathematics - Combinatorics</subject><creationdate>2024-01</creationdate><rights>http://arxiv.org/licenses/nonexclusive-distrib/1.0</rights><oa>free_for_read</oa><woscitedreferencessubscribed>false</woscitedreferencessubscribed></display><links><openurl>$$Topenurl_article</openurl><openurlfulltext>$$Topenurlfull_article</openurlfulltext><thumbnail>$$Tsyndetics_thumb_exl</thumbnail><link.rule.ids>228,230,780,885</link.rule.ids><linktorsrc>$$Uhttps://arxiv.org/abs/2401.08514$$EView_record_in_Cornell_University$$FView_record_in_$$GCornell_University$$Hfree_for_read</linktorsrc><backlink>$$Uhttps://doi.org/10.48550/arXiv.2401.08514$$DView paper in arXiv$$Hfree_for_read</backlink></links><search><creatorcontrib>Zhang, Bohang</creatorcontrib><creatorcontrib>Gai, Jingchu</creatorcontrib><creatorcontrib>Du, Yiheng</creatorcontrib><creatorcontrib>Ye, Qiwei</creatorcontrib><creatorcontrib>He, Di</creatorcontrib><creatorcontrib>Wang, Liwei</creatorcontrib><title>Beyond Weisfeiler-Lehman: A Quantitative Framework for GNN Expressiveness</title><description>Designing expressive Graph Neural Networks (GNNs) is a fundamental topic in the graph learning community. So far, GNN expressiveness has been primarily assessed via the Weisfeiler-Lehman (WL) hierarchy. However, such an expressivity measure has notable limitations: it is inherently coarse, qualitative, and may not well reflect practical requirements (e.g., the ability to encode substructures). In this paper, we introduce a unified framework for quantitatively studying the expressiveness of GNN architectures, addressing all the above limitations. Specifically, we identify a fundamental expressivity measure termed homomorphism expressivity, which quantifies the ability of GNN models to count graphs under homomorphism. Homomorphism expressivity offers a complete and practical assessment tool: the completeness enables direct expressivity comparisons between GNN models, while the practicality allows for understanding concrete GNN abilities such as subgraph counting. By examining four classes of prominent GNNs as case studies, we derive simple, unified, and elegant descriptions of their homomorphism expressivity for both invariant and equivariant settings. Our results provide novel insights into a series of previous work, unify the landscape of different subareas in the community, and settle several open questions. Empirically, extensive experiments on both synthetic and real-world tasks verify our theory, showing that the practical performance of GNN models aligns well with the proposed metric.</description><subject>Computer Science - Data Structures and Algorithms</subject><subject>Computer Science - Discrete Mathematics</subject><subject>Computer Science - Learning</subject><subject>Mathematics - Combinatorics</subject><fulltext>true</fulltext><rsrctype>article</rsrctype><creationdate>2024</creationdate><recordtype>article</recordtype><sourceid>GOX</sourceid><recordid>eNotz81OwkAUhuHZuDDoBbhybqD1zB8zdYcEkKTBmJC4bA70nDiRtmRaEe5eBFfv4ku-5BHiQUFug3PwhOkYD7m2oHIITtlbsXyhU9fW8oNizxR3lLKSPhtsn-VEvn9jO8QBh3ggOU_Y0E-XviR3SS5WKzk77hP1_Xlsz7kTN4y7nu7_OxLr-Ww9fc3Kt8VyOikzHHubUQAARudC4a0PgQ0Q1L5WuHWgmQvabIiNL7bEwF5pNmwLp7HmAFqPzUg8Xm8vlmqfYoPpVP2ZqovJ_ALT6EcP</recordid><startdate>20240116</startdate><enddate>20240116</enddate><creator>Zhang, Bohang</creator><creator>Gai, Jingchu</creator><creator>Du, Yiheng</creator><creator>Ye, Qiwei</creator><creator>He, Di</creator><creator>Wang, Liwei</creator><scope>AKY</scope><scope>AKZ</scope><scope>GOX</scope></search><sort><creationdate>20240116</creationdate><title>Beyond Weisfeiler-Lehman: A Quantitative Framework for GNN Expressiveness</title><author>Zhang, Bohang ; Gai, Jingchu ; Du, Yiheng ; Ye, Qiwei ; He, Di ; Wang, Liwei</author></sort><facets><frbrtype>5</frbrtype><frbrgroupid>cdi_FETCH-LOGICAL-a674-e8000fa558974788f30e0d7d1ac502ff9ebbef379cef0f712f3f4952adf802263</frbrgroupid><rsrctype>articles</rsrctype><prefilter>articles</prefilter><language>eng</language><creationdate>2024</creationdate><topic>Computer Science - Data Structures and Algorithms</topic><topic>Computer Science - Discrete Mathematics</topic><topic>Computer Science - Learning</topic><topic>Mathematics - Combinatorics</topic><toplevel>online_resources</toplevel><creatorcontrib>Zhang, Bohang</creatorcontrib><creatorcontrib>Gai, Jingchu</creatorcontrib><creatorcontrib>Du, Yiheng</creatorcontrib><creatorcontrib>Ye, Qiwei</creatorcontrib><creatorcontrib>He, Di</creatorcontrib><creatorcontrib>Wang, Liwei</creatorcontrib><collection>arXiv Computer Science</collection><collection>arXiv Mathematics</collection><collection>arXiv.org</collection></facets><delivery><delcategory>Remote Search Resource</delcategory><fulltext>fulltext_linktorsrc</fulltext></delivery><addata><au>Zhang, Bohang</au><au>Gai, Jingchu</au><au>Du, Yiheng</au><au>Ye, Qiwei</au><au>He, Di</au><au>Wang, Liwei</au><format>journal</format><genre>article</genre><ristype>JOUR</ristype><atitle>Beyond Weisfeiler-Lehman: A Quantitative Framework for GNN Expressiveness</atitle><date>2024-01-16</date><risdate>2024</risdate><abstract>Designing expressive Graph Neural Networks (GNNs) is a fundamental topic in the graph learning community. So far, GNN expressiveness has been primarily assessed via the Weisfeiler-Lehman (WL) hierarchy. However, such an expressivity measure has notable limitations: it is inherently coarse, qualitative, and may not well reflect practical requirements (e.g., the ability to encode substructures). In this paper, we introduce a unified framework for quantitatively studying the expressiveness of GNN architectures, addressing all the above limitations. Specifically, we identify a fundamental expressivity measure termed homomorphism expressivity, which quantifies the ability of GNN models to count graphs under homomorphism. Homomorphism expressivity offers a complete and practical assessment tool: the completeness enables direct expressivity comparisons between GNN models, while the practicality allows for understanding concrete GNN abilities such as subgraph counting. By examining four classes of prominent GNNs as case studies, we derive simple, unified, and elegant descriptions of their homomorphism expressivity for both invariant and equivariant settings. Our results provide novel insights into a series of previous work, unify the landscape of different subareas in the community, and settle several open questions. Empirically, extensive experiments on both synthetic and real-world tasks verify our theory, showing that the practical performance of GNN models aligns well with the proposed metric.</abstract><doi>10.48550/arxiv.2401.08514</doi><oa>free_for_read</oa></addata></record>
fulltext fulltext_linktorsrc
identifier DOI: 10.48550/arxiv.2401.08514
ispartof
issn
language eng
recordid cdi_arxiv_primary_2401_08514
source arXiv.org
subjects Computer Science - Data Structures and Algorithms
Computer Science - Discrete Mathematics
Computer Science - Learning
Mathematics - Combinatorics
title Beyond Weisfeiler-Lehman: A Quantitative Framework for GNN Expressiveness
url https://sfx.bib-bvb.de/sfx_tum?ctx_ver=Z39.88-2004&ctx_enc=info:ofi/enc:UTF-8&ctx_tim=2024-12-24T22%3A03%3A57IST&url_ver=Z39.88-2004&url_ctx_fmt=infofi/fmt:kev:mtx:ctx&rfr_id=info:sid/primo.exlibrisgroup.com:primo3-Article-arxiv_GOX&rft_val_fmt=info:ofi/fmt:kev:mtx:journal&rft.genre=article&rft.atitle=Beyond%20Weisfeiler-Lehman:%20A%20Quantitative%20Framework%20for%20GNN%20Expressiveness&rft.au=Zhang,%20Bohang&rft.date=2024-01-16&rft_id=info:doi/10.48550/arxiv.2401.08514&rft_dat=%3Carxiv_GOX%3E2401_08514%3C/arxiv_GOX%3E%3Curl%3E%3C/url%3E&disable_directlink=true&sfx.directlink=off&sfx.report_link=0&rft_id=info:oai/&rft_id=info:pmid/&rfr_iscdi=true