{"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/deeper-insights-into-graph-convolutional","title":"Deeper Insights into Graph Convolutional Networks for Semi-Supervised Learning","arxiv_id":"1801.07606","date":"2018-01-22","proceeding":null,"authors":["Qimai Li","Zhichao Han","Xiao-Ming Wu"],"abstract":"Many interesting problems in machine learning are being revisited with new\ndeep learning tools. For graph-based semisupervised learning, a recent\nimportant development is graph convolutional networks (GCNs), which nicely\nintegrate local vertex features and graph topology in the convolutional layers.\nAlthough the GCN model compares favorably with other state-of-the-art methods,\nits mechanisms are not clear and it still requires a considerable amount of\nlabeled data for validation and model selection. In this paper, we develop\ndeeper insights into the GCN model and address its fundamental limits. First,\nwe show that the graph convolution of the GCN model is actually a special form\nof Laplacian smoothing, which is the key reason why GCNs work, but it also\nbrings potential concerns of over-smoothing with many convolutional layers.\nSecond, to overcome the limits of the GCN model with shallow architectures, we\npropose both co-training and self-training approaches to train GCNs. Our\napproaches significantly improve GCNs in learning with very few labels, and\nexempt them from requiring additional labels for validation. Extensive\nexperiments on benchmarks have verified our theory and proposals.","url_abs":"http://arxiv.org/abs/1801.07606v1","url_pdf":"http://arxiv.org/pdf/1801.07606v1.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":"deeper-insights-into-graph-convolutional","repo_url":"https://github.com/liqimai/gcn","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"tf","reach":{"status":"unanswered"}}],"tasks":[{"task_slug":"model-selection","task_name":"Model Selection"},{"task_slug":"node-classification","task_name":"Node Classification"}],"methods":[{"method_slug":"convolution","method_name":"Convolution"},{"method_slug":"gcn","method_name":"GCN"},{"method_slug":"graph-convolutional-networks","method_name":"Graph Convolutional Networks"}],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/node-classification-on-brazil-air-traffic","task":"Node Classification","dataset":"Brazil Air-Traffic","model":"Union (Li et al., 2018)","rank_in_archive_order":4,"of":7,"metrics":{"Accuracy":"0.466"},"uses_additional_data":false},{"leaderboard":"/sota/node-classification-on-brazil-air-traffic","task":"Node Classification","dataset":"Brazil Air-Traffic","model":"Intersection (Li et al., 2018)","rank_in_archive_order":5,"of":7,"metrics":{"Accuracy":"0.459"},"uses_additional_data":false},{"leaderboard":"/sota/node-classification-on-europe-air-traffic","task":"Node Classification","dataset":"Europe Air-Traffic","model":"Intersection (Li et al., 2018)","rank_in_archive_order":4,"of":7,"metrics":{"Accuracy":"44.3"},"uses_additional_data":false},{"leaderboard":"/sota/node-classification-on-facebook","task":"Node Classification","dataset":"Facebook","model":"Intersection (Li et al., 2018)","rank_in_archive_order":6,"of":8,"metrics":{"Accuracy":"59.8"},"uses_additional_data":false},{"leaderboard":"/sota/node-classification-on-flickr","task":"Node Classification","dataset":"Flickr","model":"Intersection (Li et al., 2018)","rank_in_archive_order":5,"of":8,"metrics":{"Accuracy":"0.557"},"uses_additional_data":false},{"leaderboard":"/sota/node-classification-on-usa-air-traffic","task":"Node Classification","dataset":"USA Air-Traffic","model":"Union (Li et al., 2018)","rank_in_archive_order":5,"of":7,"metrics":{"Accuracy":"58.2"},"uses_additional_data":false},{"leaderboard":"/sota/node-classification-on-usa-air-traffic","task":"Node Classification","dataset":"USA Air-Traffic","model":"Intersection (Li et al., 2018)","rank_in_archive_order":6,"of":7,"metrics":{"Accuracy":"57.3"},"uses_additional_data":false},{"leaderboard":"/sota/node-classification-on-wiki-vote","task":"Node Classification","dataset":"Wiki-Vote","model":"Union (Li et al., 2018)","rank_in_archive_order":4,"of":6,"metrics":{"Accuracy":"46.3"},"uses_additional_data":false}],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1801.07606","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}