Generalization and Representational Limits of Graph Neural Networks

被引:0
|
作者
Garg, Vikas K. [1 ]
Jegelka, Stefanie [1 ]
Jaakkola, Tommi [1 ]
机构
[1] MIT, CSAIL, 77 Massachusetts Ave, Cambridge, MA 02139 USA
基金
美国国家科学基金会;
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 address two fundamental questions about graph neural networks (GNNs). First, we prove that several important graph properties, e.g., shortest/longest cycle, diameter, or certain motifs, cannot be computed by GNNs that rely entirely on local information. Such GNNs include the standard message passing models, and more powerful variants that exploit local graph structure (e.g., via relative orientation of messages, or local port ordering) to distinguish neighbors of each node. Our treatment includes a novel graph-theoretic formalism. Second, we provide the first data dependent generalization bounds for message passing GNNs. This analysis explicitly accounts for the local permutation invariance of GNNs. Our bounds are much tighter than existing VC-dimension based guarantees for GNNs, and are comparable to Rademacher bounds for recurrent neural networks.
引用
收藏
页数:12
相关论文
共 50 条
  • [1] Generalization and Representational Limits of Graph Neural Networks
    Garg, Vikas K.
    Jegelka, Stefanie
    Jaakkola, Tommi
    25TH AMERICAS CONFERENCE ON INFORMATION SYSTEMS (AMCIS 2019), 2019,
  • [2] Generalization limits of Graph Neural Networks in identity effects learning
    D'Inverno, Giuseppe Alessio
    Brugiapaglia, Simone
    Ravanelli, Mirco
    NEURAL NETWORKS, 2025, 181
  • [3] A Generalization of Recurrent Neural Networks for Graph Embedding
    Han, Xiao
    Zhang, Chunhong
    Guo, Chenchen
    Ji, Yang
    ADVANCES IN KNOWLEDGE DISCOVERY AND DATA MINING, PAKDD 2018, PT II, 2018, 10938 : 247 - 259
  • [4] Stability and Generalization of Graph Convolutional Neural Networks
    Verma, Saurabh
    Zhang, Zhi-Li
    KDD'19: PROCEEDINGS OF THE 25TH ACM SIGKDD INTERNATIONAL CONFERENCCE ON KNOWLEDGE DISCOVERY AND DATA MINING, 2019, : 1539 - 1548
  • [5] Subgroup Generalization and Fairness of Graph Neural Networks
    Ma, Jiaqi
    Deng, Junwei
    Mei, Qiaozhu
    ADVANCES IN NEURAL INFORMATION PROCESSING SYSTEMS 34 (NEURIPS 2021), 2021,
  • [6] Understanding Attention and Generalization in Graph Neural Networks
    Knyazev, Boris
    Taylor, Graham W.
    Amer, Mohamed R.
    ADVANCES IN NEURAL INFORMATION PROCESSING SYSTEMS 32 (NIPS 2019), 2019, 32
  • [7] On the Topology Awareness and Generalization Performance of Graph Neural Networks
    Su, Junwei
    Wu, Chuan
    COMPUTER VISION - ECCV 2024, PT LXXXIV, 2025, 15142 : 73 - 89
  • [8] Breaking the Limits of Message Passing Graph Neural Networks
    Balcilar, Muhammet
    Heroux, Pierre
    Gauzere, Benoit
    Vasseur, Pascal
    Adam, Sebastien
    Honeine, Paul
    INTERNATIONAL CONFERENCE ON MACHINE LEARNING, VOL 139, 2021, 139
  • [9] Adversarial Weight Perturbation Improves Generalization in Graph Neural Networks
    Wu, Yihan
    Bojchevski, Aleksandar
    Huang, Heng
    THIRTY-SEVENTH AAAI CONFERENCE ON ARTIFICIAL INTELLIGENCE, VOL 37 NO 9, 2023, : 10417 - 10425
  • [10] Challenging the generalization capabilities of Graph Neural Networks for network modeling
    Suarez-Varela, Jose
    Carol-Bosch, Sergi
    Rusek, Krzysztof
    Almasan, Paul
    Arias, Marta
    Barlet-Ros, Pere
    Cabellos-Aparicio, Albert
    PROCEEDINGS OF THE 2019 ACM SIGCOMM CONFERENCE POSTERS AND DEMOS (SIGCOMM '19), 2019, : 114 - 115