Abstract:Over-smoothing has emerged as a severe obstacle to node classification with message passing based graph convolutional networks (GCNs). Classification performance dramatically deteriorates for deep GCNs, as message passing over the observed noisy graph topology cannot adequately propagate intra-class information and over-mix the features of nodes from different communities (classes). Existing topology optimization methods for GCNs cannot sufficiently exploit the underlying ground-truth community structure to distinguish nodes from different communities. In this paper, we propose a novel method, termed EM-GCN, to address this issue by employing the Expectation Maximization (EM) algorithm to simultaneously achieve community-enhanced topology optimization and learn desirable node representations for classification. EM-GCN represents the underlying community structure with a latent adjacency matrix parameterized by an assortative-constrained stochastic block model, and consequently, explicitly enhances intra-class connection and suppresses inter-class interaction in the observed noisy graph. In the inference procedure (E-step), a graph inference model shared across all the node pairs is learned from node embeddings to approximate the posterior distribution of the latent adjacency matrix and optimize the graph topology. In the learning procedure (M-step), node representations are learned using GCNs based on the refined graph topology for the downstream classification task. EM-GCN is a general and flexible method that leverages approximate posterior and arbitrary GCNs for overcoming over-smoothing with topology optimization. Experimental results on synthetic and real-world datasets demonstrate that EM-GCN outperforms existing strategies for tackling over-smoothing and optimizing graph topology in node classification.

Degree-Specific Topology Learning for Graph Convolutional Network

Graph Learning in Low Dimensional Space for Graph Convolutional Networks

Contrastive Graph Convolutional Networks with Degree-specific Augmentation

Topology Optimization Based Graph Convolutional Network.

Investigating and Mitigating Degree-Related Biases in Graph Convolutional Networks

Graph Topology Noise Aware Learning by Feature Clustering and Pseudo-labels Generator

Graph Convolutional Network with elastic topology

Generalization Guarantee of Training Graph Convolutional Networks with Graph Topology Sampling

Tackling Over-Smoothing for General Graph Convolutional Networks

DropEdge: Towards Deep Graph Convolutional Networks on Node Classification.

Topological based classification using graph convolutional networks

Neighborhood Pattern Is Crucial for Graph Convolutional Networks Performing Node Classification

Tackling Over-Smoothing in Graph Convolutional Networks With EM-Based Joint Topology Optimization and Node Classification

Dropedge: towards deep graph convolu-

Semi-Supervised Classification of Graph Convolutional Networks with Laplacian Rank Constraints

High-order Graph Convolutional Networks for Semi-Supervised Classification

Deeper Insights Into Graph Convolutional Networks for Semi-Supervised Learning

WGCN: Graph Convolutional Networks with Weighted Structural Features

Learnable Graph Convolutional Network with Semisupervised Graph Information Bottleneck.

Revisiting Graph Convolutional Network on Semi-Supervised Node Classification from an Optimization Perspective

Graph convolutional networks of reconstructed graph structure with constrained Laplacian rank