Simple and Deep Graph Convolutional Networks
Ming Chen Zhewei Wei Zengfeng Huang Bolin Ding Yaliang Li
Abstract
Graph convolutional networks (GCNs) are a powerful deep learning approach for graph-structured data. Recently, GCNs and subsequent variants have shown superior performance in various application areas on real-world datasets. Despite their success, most of the current GCN models are shallow, due to the over-smoothing problem. In this paper, we study the problem of designing and analyzing deep graph convolutional networks. We propose the GCNII, an extension of the vanilla GCN model with two simple yet effective techniques: Initial residual and Identity mapping. We provide theoretical and empirical evidence that the two techniques effectively relieves the problem of over-smoothing. Our experiments show that the deep GCNII model outperforms the state-of-the-art methods on various semi- and full-supervised tasks. Code is available at https://github.com/chennnM/GCNII.
中文速览
图神经网络（GCN）在实际应用中表现出色，但长期以来只能做两三层的浅层模型——层数一多，各节点的表示就会越来越相似、失去区分度，这就是"过平滑（over-smoothing）"问题。作者提出了 GCNII 模型，在标准 GCN 的基础上加入两个简单改动：一是"初始残差连接"，让每一层都保留一部分来自输入层的原始特征；二是"恒等映射"，在每层的权重矩阵上叠加一个单位矩阵，保证深层模型至少不比浅层差。理论上，作者证明了这两项改动使 GCNII 能够表达任意系数的多项式谱滤波器，从根本上突破了普通 GCN 系数固定的瓶颈；实验则证实了高度数节点更容易过平滑，而 GCNII 在半监督和全监督的节点分类任务上均超过了此前最优方法。这项工作的意义在于首次证明"深度"对图神经网络确实是一种资源而非负担，为后续设计更深、更强的图模型提供了理论依据和实践范本。
原文 arXiv:2007.02133;中英对照 + 大白话阅读 https://aha.fim.ai/paper/2007.02133v1