{"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/soft-filter-pruning-for-accelerating-deep","title":"Soft Filter Pruning for Accelerating Deep Convolutional Neural Networks","arxiv_id":"1808.06866","date":"2018-08-21","proceeding":null,"authors":["Yang He","Guoliang Kang","Xuanyi Dong","Yanwei Fu","Yi Yang"],"abstract":"This paper proposed a Soft Filter Pruning (SFP) method to accelerate the\ninference procedure of deep Convolutional Neural Networks (CNNs). Specifically,\nthe proposed SFP enables the pruned filters to be updated when training the\nmodel after pruning. SFP has two advantages over previous works: (1) Larger\nmodel capacity. Updating previously pruned filters provides our approach with\nlarger optimization space than fixing the filters to zero. Therefore, the\nnetwork trained by our method has a larger model capacity to learn from the\ntraining data. (2) Less dependence on the pre-trained model. Large capacity\nenables SFP to train from scratch and prune the model simultaneously. In\ncontrast, previous filter pruning methods should be conducted on the basis of\nthe pre-trained model to guarantee their performance. Empirically, SFP from\nscratch outperforms the previous filter pruning methods. Moreover, our approach\nhas been demonstrated effective for many advanced CNN architectures. Notably,\non ILSCRC-2012, SFP reduces more than 42% FLOPs on ResNet-101 with even 0.2%\ntop-5 accuracy improvement, which has advanced the state-of-the-art. Code is\npublicly available on GitHub: https://github.com/he-y/soft-filter-pruning","url_abs":"http://arxiv.org/abs/1808.06866v1","url_pdf":"http://arxiv.org/pdf/1808.06866v1.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":"soft-filter-pruning-for-accelerating-deep","repo_url":"https://github.com/he-y/soft-filter-pruning","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":0,"framework":"pytorch","reach":{"status":"unanswered"}},{"paper_slug":"soft-filter-pruning-for-accelerating-deep","repo_url":"https://github.com/AlumLuther/PruningFilters","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"unanswered"}},{"paper_slug":"soft-filter-pruning-for-accelerating-deep","repo_url":"https://github.com/EkdeepSLubana/OrthoReg","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":null},{"paper_slug":"soft-filter-pruning-for-accelerating-deep","repo_url":"https://github.com/Gideon0805/Tensorflow1.15-Model-Pruning","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"tf","reach":null},{"paper_slug":"soft-filter-pruning-for-accelerating-deep","repo_url":"https://github.com/Gideon0805/Tensorflow_Model_Pruning","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"tf","reach":{"status":"unanswered"}},{"paper_slug":"soft-filter-pruning-for-accelerating-deep","repo_url":"https://github.com/arturjordao/PruningNeuralNetworks","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"tf","reach":{"status":"unanswered"}}],"tasks":[],"methods":[{"method_slug":"pruning","method_name":"Pruning"}],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1808.06866","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}