{"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/meprop-sparsified-back-propagation-for","title":"meProp: Sparsified Back Propagation for Accelerated Deep Learning with Reduced Overfitting","arxiv_id":"1706.06197","date":"2017-06-19","proceeding":"ICML 2017 8","authors":["Xu Sun","Xuancheng Ren","Shuming Ma","Houfeng Wang"],"abstract":"We propose a simple yet effective technique for neural network learning. The\nforward propagation is computed as usual. In back propagation, only a small\nsubset of the full gradient is computed to update the model parameters. The\ngradient vectors are sparsified in such a way that only the top-$k$ elements\n(in terms of magnitude) are kept. As a result, only $k$ rows or columns\n(depending on the layout) of the weight matrix are modified, leading to a\nlinear reduction ($k$ divided by the vector dimension) in the computational\ncost. Surprisingly, experimental results demonstrate that we can update only\n1-4% of the weights at each back propagation pass. This does not result in a\nlarger number of training iterations. More interestingly, the accuracy of the\nresulting models is actually improved rather than degraded, and a detailed\nanalysis is given. The code is available at https://github.com/lancopku/meProp","url_abs":"http://arxiv.org/abs/1706.06197v5","url_pdf":"http://arxiv.org/pdf/1706.06197v5.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":"meprop-sparsified-back-propagation-for","repo_url":"https://github.com/jklj077/meProp","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"unanswered"}},{"paper_slug":"meprop-sparsified-back-propagation-for","repo_url":"https://github.com/lancopku/meProp","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"unanswered"}}],"tasks":[{"task_slug":"deep-learning","task_name":"Deep Learning"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"syntology_url":"https://syntology.ai/paper/1706.06197","atlas_url":"https://app.syntology.ai/?focus=1706.06197","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}