Graph Neural Networks Exponentially Lose Expressive Power for Node Classification

Abstract

Graph Neural Networks (graph NNs) are a promising deep learning approach for analyzing graph-structured data. However, it is known that they do not improve (or sometimes worsen) their predictive performance as we pile up many layers and add non-lineality. To tackle this problem, we investigate the expressive power of graph NNs via their asymptotic behaviors as the layer size tends to infinity. Our strategy is to generalize the forward propagation of a Graph Convolutional Network (GCN), which is a popular graph NN variant, as a specific dynamical system. In the case of a GCN, we show that when its weights satisfy the conditions determined by the spectra of the (augmented) normalized Laplacian, its output exponentially approaches the set of signals that carry information of the connected components and node degrees only for distinguishing nodes. Our theory enables us to relate the expressive power of GCNs with the topological information of the underlying graphs inherent in the graph spectra. To demonstrate this, we characterize the asymptotic behavior of GCNs on the Erdős -- Rényi graph. We show that when the Erdős -- Rényi graph is sufficiently dense and large, a broad range of GCNs on it suffers from the "information loss" in the limit of infinite layers with high probability. Based on the theory, we provide a principled guideline for weight normalization of graph NNs. We experimentally confirm that the proposed weight scaling enhances the predictive performance of GCNs in real data. Code is available at https://github.com/delta2323/gnn-asymptotics.

9 pages, Supplemental material 28 pages. Accepted in International Conference on Learning Representations (ICLR) 2020

References in corpus (11)

• Neural Message Passing for Quantum Chemistry
• Reconciling modern machine learning practice and the bias-variance trade-off
• Deep Convolutional Networks on Graph-Structured Data
• Simplifying Graph Convolutional Networks
• Representation Learning on Graphs with Jumping Knowledge Networks
• Revisiting Graph Neural Networks: All We Have is Low-Pass Filters
• The generalization error of random features regression: Precise asymptotics and double descent curve
• Dynamical Isometry and a Mean Field Theory of RNNs: Gating Enables Signal Propagation in Recurrent Neural Networks
• Approximation and Non-parametric Estimation of ResNet-type Convolutional Neural Networks
• Understanding Generalization and Optimization Performance of Deep CNNs
• Semi-supervised learning of hierarchical representations of molecules using neural message passing

Cited by in corpus (20)

• Graph Signal Processing and Deep Learning: Convolution, Pooling, and Topology
• When Does Self-Supervision Help Graph Convolutional Networks?
• DropMessage: Unifying Random Dropping for Graph Neural Networks
• Model Degradation Hinders Deep Graph Neural Networks
• Propagation Tree Is Not Deep: Adaptive Graph Contrastive Learning Approach for Rumor Detection
• DenseGNN: universal and scalable deeper graph neural networks for high-performance property prediction in crystals and molecules
• Graph Neural Networks: Methods, Applications, and Opportunities
• Feature Overcorrelation in Deep Graph Neural Networks: A New Perspective
• Structure-Aware DropEdge Towards Deep Graph Convolutional Networks
• GCNH: A Simple Method For Representation Learning On Heterophilous Graphs
• Beyond Graph Convolutional Network: An Interpretable Regularizer-centered Optimization Framework
• AnchorGAE: General Data Clustering via $O(n)$ Bipartite Graph Convolution
• Multiresolution Graph Transformers and Wavelet Positional Encoding for Learning Hierarchical Structures
• A Multidimensional Graph Fourier Transformation Neural Network for Vehicle Trajectory Prediction
• Transient Chaos in BERT
• Weisfeiler and Lehman Go Paths: Learning Topological Features via Path Complexes
• Clarify Confused Nodes via Separated Learning
• DeepHGCN: Toward Deeper Hyperbolic Graph Convolutional Networks
• Sheaf4Rec: Sheaf Neural Networks for Graph-based Recommender Systems
• Multi-Level Attention Pooling for Graph Neural Networks: Unifying Graph Representations with Multiple Localities