{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/residual-gated-graph-convnets","title":"Residual Gated Graph ConvNets","arxiv_id":"1711.07553","date":"2017-11-20","proceeding":"ICLR 2018 1","authors":["Xavier Bresson","Thomas Laurent"],"abstract":"Graph-structured data such as social networks, functional brain networks,\ngene regulatory networks, communications networks have brought the interest in\ngeneralizing deep learning techniques to graph domains. In this paper, we are\ninterested to design neural networks for graphs with variable length in order\nto solve learning problems such as vertex classification, graph classification,\ngraph regression, and graph generative tasks. Most existing works have focused\non recurrent neural networks (RNNs) to learn meaningful representations of\ngraphs, and more recently new convolutional neural networks (ConvNets) have\nbeen introduced. In this work, we want to compare rigorously these two\nfundamental families of architectures to solve graph learning tasks. We review\nexisting graph RNN and ConvNet architectures, and propose natural extension of\nLSTM and ConvNet to graphs with arbitrary size. Then, we design a set of\nanalytically controlled experiments on two basic graph problems, i.e. subgraph\nmatching and graph clustering, to test the different architectures. Numerical\nresults show that the proposed graph ConvNets are 3-17% more accurate and\n1.5-4x faster than graph RNNs. Graph ConvNets are also 36% more accurate than\nvariational (non-learning) techniques. Finally, the most effective graph\nConvNet architecture uses gated edges and residuality. Residuality plays an\nessential role to learn multi-layer architectures as they provide a 10% gain of\nperformance.","url_abs":"http://arxiv.org/abs/1711.07553v2","url_pdf":"http://arxiv.org/pdf/1711.07553v2.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"residual-gated-graph-convnets","repo_url":"https://github.com/xbresson/spatial_graph_convnets","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"unanswered"}}],"tasks":[{"task_slug":"clustering","task_name":"Clustering"},{"task_slug":"classification","task_name":"General Classification"},{"task_slug":"graph-classification","task_name":"Graph Classification"},{"task_slug":"graph-clustering","task_name":"Graph Clustering"},{"task_slug":"graph-learning","task_name":"Graph Learning"},{"task_slug":"graph-regression","task_name":"Graph Regression"},{"task_slug":"node-classification","task_name":"Node Classification"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/graph-classification-on-cifar10-100k","task":"Graph Classification","dataset":"CIFAR10 100k","model":"GatedGCN","rank_in_archive_order":14,"of":20,"metrics":{"Accuracy (%)":"69.37"},"uses_additional_data":false},{"leaderboard":"/sota/graph-regression-on-zinc-500k","task":"Graph Regression","dataset":"ZINC-500k","model":"GatedGCN","rank_in_archive_order":30,"of":36,"metrics":{"MAE":"0.282 "},"uses_additional_data":false},{"leaderboard":"/sota/node-classification-on-pattern-100k","task":"Node Classification","dataset":"PATTERN 100k","model":"GatedGCN","rank_in_archive_order":7,"of":9,"metrics":{"Accuracy (%)":"84.480"},"uses_additional_data":false}],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1711.07553","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}