arXiv (Cornell University) · 2020 · 402 citations · 0 references
ComputingComputer ScienceComputer VisionKnowledge NetworksMachine LearningDeep LearningGraph AlgorithmsGraph AnalyticsGraph TheoryGraph-based LearningRepresentation LearningBenchmark DatasetsConvolutional Neural NetworksGraph Convolutional NetworksGraph Neural NetworksGraph Representation LearningSemi-supervised Graph Neural NetworksSemi-supervised Learning On Graphs
Graph convolutional networks (GCNs) are a powerful deep learning approach for graph‑structured data that have achieved superior performance on real‑world datasets, yet most current models remain shallow due to over‑smoothing. This work investigates how to design and analyze deep graph convolutional networks. The authors introduce GCNII, an extension of vanilla GCN that incorporates initial residual connections and identity mapping to enable deeper architectures. Theoretical analysis and experiments demonstrate that these techniques alleviate over‑smoothing and that deep GCNII outperforms state‑of‑the‑art methods on semi‑ and fully supervised graph tasks. Code is available at the provided URL.
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 {\em 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: {\em Initial residual} and {\em 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 this https URL .