What functions can Graph Neural Networks compute on random graphs? The role of Positional Encoding

被引:0
|
作者
Keriven, Nicolas [1 ]
Vaiter, Samuel [2 ]
机构
[1] CNRS, IRISA, Rennes, France
[2] CNRS, LJAD, Nice, France
关键词
D O I
暂无
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
We aim to deepen the theoretical understanding of Graph Neural Networks (GNNs) on large graphs, with a focus on their expressive power. Existing analyses relate this notion to the graph isomorphism problem, which is mostly relevant for graphs of small sizes, or studied graph classification or regression tasks, while prediction tasks on nodes are far more relevant on large graphs. Recently, several works showed that, on very general random graphs models, GNNs converge to certains functions as the number of nodes grows. In this paper, we provide a more complete and intuitive description of the function space generated by equivariant GNNs for node-tasks, through general notions of convergence that encompass several previous examples. We emphasize the role of input node features, and study the impact of node Positional Encodings (PEs), a recent line of work that has been shown to yield state-of-the-art results in practice. Through the study of several examples of PEs on large random graphs, we extend previously known universality results to significantly more general models. Our theoretical results hint at some normalization tricks, which is shown numerically to have a positive impact on GNN generalization on synthetic and real data. Our proofs contain new concentration inequalities of independent interest.
引用
收藏
页数:27
相关论文
共 50 条
  • [11] Instant Graph Neural Networks for Dynamic Graphs
    Zheng, Yanping
    Wang, Hanzhi
    Wei, Zhewei
    Liu, Jiajun
    Wang, Sibo
    PROCEEDINGS OF THE 28TH ACM SIGKDD CONFERENCE ON KNOWLEDGE DISCOVERY AND DATA MINING, KDD 2022, 2022, : 2605 - 2615
  • [12] Ranking Attack Graphs with Graph Neural Networks
    Lu, Liang
    Safavi-Naini, Rei
    Hagenbuchner, Markus
    Susilo, Willy
    Horton, Jeffrey
    Yong, Sweah Liang
    Tsoi, Ah Chung
    INFORMATION SECURITY PRACTICE AND EXPERIENCE, PROCEEDINGS: 5TH INTERNATIONAL CONFERENCE, ISPEC 2009, 2009, 5451 : 345 - +
  • [13] Sheaf-based Positional Encodings for Graph Neural Networks
    He, Yu
    Bodnar, Cristian
    Lio, Pietro
    NEURIPS WORKSHOP ON SYMMETRY AND GEOMETRY IN NEURAL REPRESENTATIONS, 2023, 228 : 1 - 18
  • [14] Neural Graph Learning: Training Neural Networks Using Graphs
    Bui, Thang D.
    Ravi, Sujith
    Ramavajjala, Vivek
    WSDM'18: PROCEEDINGS OF THE ELEVENTH ACM INTERNATIONAL CONFERENCE ON WEB SEARCH AND DATA MINING, 2018, : 64 - 71
  • [15] Analogue neural networks on correlated random graphs
    Agliari, Elena
    Asti, Lorenzo
    Barra, Adriano
    Burioni, Raffaella
    Uguzzoni, Guido
    JOURNAL OF PHYSICS A-MATHEMATICAL AND THEORETICAL, 2012, 45 (36)
  • [16] A Comprehensive Survey of Graph Neural Networks for Knowledge Graphs
    Ye, Zi
    Kumar, Yogan Jaya
    Sing, Goh Ong
    Song, Fengyan
    Wang, Junsong
    IEEE ACCESS, 2022, 10 : 75729 - 75741
  • [17] Efficient Training of Graph Neural Networks on Large Graphs
    Shen, Yanyan
    Chen, Lei
    Fang, Jingzhi
    Zhang, Xin
    Gao, Shihong
    Yin, Hongbo
    PROCEEDINGS OF THE VLDB ENDOWMENT, 2024, 17 (12): : 4237 - 4240
  • [18] Convergence of Graph Neural Networks on Relatively Sparse Graphs
    Wang, Zhiyang
    Ruiz, Luana
    Ribeiro, Alejandro
    FIFTY-SEVENTH ASILOMAR CONFERENCE ON SIGNALS, SYSTEMS & COMPUTERS, IEEECONF, 2023, : 566 - 572
  • [19] How hard is to distinguish graphs with graph neural networks?
    Loukas, Andreas
    ADVANCES IN NEURAL INFORMATION PROCESSING SYSTEMS 33, NEURIPS 2020, 2020, 33
  • [20] Decoupled Graph Neural Networks for Large Dynamic Graphs
    Zheng, Yanping
    Wei, Zhewei
    Liu, Jiajun
    PROCEEDINGS OF THE VLDB ENDOWMENT, 2023, 16 (09): : 2239 - 2247