{"id":20663758,"url":"https://github.com/vita-group/gradientgcn","last_synced_at":"2025-08-14T07:36:08.872Z","repository":{"id":77794991,"uuid":"543334847","full_name":"VITA-Group/GradientGCN","owner":"VITA-Group","description":"[NeurIPS 2022] Old can be Gold: Better Gradient Flow can Make Vanilla-GCNs Great Again by Ajay Jaiswal*, Peihao Wang*, Tianlong Chen, Justin F Rousseau, Ying Ding, Zhangyang Wang","archived":false,"fork":false,"pushed_at":"2022-11-25T03:46:00.000Z","size":83,"stargazers_count":8,"open_issues_count":1,"forks_count":0,"subscribers_count":11,"default_branch":"main","last_synced_at":"2025-03-29T09:42:01.572Z","etag":null,"topics":["deep-gcns","gradient-flow","graph-neural-networks","initialization"],"latest_commit_sha":null,"homepage":"","language":"Python","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":null,"status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/VITA-Group.png","metadata":{"files":{"readme":"README.md","changelog":null,"contributing":null,"funding":null,"license":null,"code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":null,"support":null,"governance":null,"roadmap":null,"authors":null,"dei":null,"publiccode":null,"codemeta":null}},"created_at":"2022-09-29T22:01:57.000Z","updated_at":"2023-06-22T02:53:28.000Z","dependencies_parsed_at":"2023-03-08T16:45:58.108Z","dependency_job_id":null,"html_url":"https://github.com/VITA-Group/GradientGCN","commit_stats":null,"previous_names":[],"tags_count":0,"template":false,"template_full_name":null,"repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/VITA-Group%2FGradientGCN","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/VITA-Group%2FGradientGCN/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/VITA-Group%2FGradientGCN/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/VITA-Group%2FGradientGCN/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/VITA-Group","download_url":"https://codeload.github.com/VITA-Group/GradientGCN/tar.gz/refs/heads/main","host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":249731218,"owners_count":21317341,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2022-07-04T15:15:14.044Z","host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":["deep-gcns","gradient-flow","graph-neural-networks","initialization"],"created_at":"2024-11-16T19:19:42.028Z","updated_at":"2025-04-19T15:55:59.917Z","avatar_url":"https://github.com/VITA-Group.png","language":"Python","funding_links":[],"categories":[],"sub_categories":[],"readme":"# Old can be Gold: Better Gradient Flow can make Vanilla-GCNs Great Again\r\n[![License: MIT](https://img.shields.io/badge/License-MIT-green.svg)](https://opensource.org/licenses/MIT)\r\n\r\nhttps://arxiv.org/abs/2210.08122\r\n\r\n## Abstract\r\n\r\nDespite the enormous success of Graph Convolutional Networks (GCNs) in mod-\r\nelling graph-structured data, most of the current GCNs are shallow due to the\r\nnotoriously challenging problems of over-smoothening and information squashing\r\nalong with conventional difficulty caused by vanishing gradients and over-fitting.\r\nPrevious works have been primarily focused on the study of over-smoothening and\r\nover-squashing phenomenon in training deep GCNs. Surprisingly, in comparison\r\nwith CNNs/RNNs, very limited attention has been given towards understanding\r\nhow healthy gradient flow can benefit the trainability of deep GCNs. In this paper,\r\nfirstly, we provide a new perspective of gradient flow to understand the substandard\r\nperformance of deep GCNs and hypothesize that by facilitating healthy gradient\r\nflow, we can significantly improve their trainability, as well as achieve state-of-the-\r\nart (SOTA) level performance from vanilla-GCNs [1]. Next, we argue that blindly\r\nadopting the Glorot initialization for GCNs is not optimal, and derive a topology-\r\naware isometric initialization scheme for vanilla-GCNs based on the principles\r\nof isometry. Additionally, contrary to ad-hoc addition of skip-connections, we\r\npropose to use gradient-guided dynamic rewiring of vanilla-GCNs with skip-\r\nconnections. Our dynamic rewiring method uses the gradient flow within each\r\nlayer during training to introduce skip-connections on-demand basis. We provide\r\nextensive empirical evidence across multiple datasets that our methods improves\r\ngradient flow in deep vanilla-GCNs and significantly boost their performance to\r\ncomfortably compete and outperform many fancy state-of-the-art methods. \r\n\r\n![image](https://user-images.githubusercontent.com/6660499/193684200-ef091d81-cb91-4496-8e6e-d7e297c573e1.png)\r\n\r\n## Benefits of our proposed techniques\r\n\r\n![image](https://user-images.githubusercontent.com/6660499/193684330-c1762f74-6fb6-478d-b305-b02d145f8fcb.png)\r\n\r\n![image](https://user-images.githubusercontent.com/6660499/193684419-42a18f00-6386-4439-ab3a-6685f1537a43.png)\r\n\r\n\r\n![image](https://user-images.githubusercontent.com/6660499/193684907-10f6d567-4922-4677-987c-1bffb6111658.png)\r\n\r\n\r\n![image](https://user-images.githubusercontent.com/6660499/193684522-455cf185-b410-453f-81a9-1be0c621c25c.png)\r\n\r\nIf you find our work helpful in your research, please cite our paper\r\n\r\n## Citation\r\n\r\nIf you find our code implementation helpful for your own resarch or work, please cite our paper.\r\n```\r\n@inproceedings{Jaiswal22GradientGCN,\r\n  title={Old can be Gold: Better Gradient Flow can make Vanilla-GCNs Great Again},\r\n  author={Ajay Jaiswal, Peihao Wang, Tianlong Chen, Justin F Rousseau, Ying Ding, Zhangyang Wang},\r\n  booktitle={NeurIPS 2022},\r\n  year={2022}\r\n}\r\n```\r\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fvita-group%2Fgradientgcn","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Fvita-group%2Fgradientgcn","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fvita-group%2Fgradientgcn/lists"}