{"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/data-driven-sparse-structure-selection-for","title":"Data-Driven Sparse Structure Selection for Deep Neural Networks","arxiv_id":"1707.01213","date":"2017-07-05","proceeding":"ECCV 2018 9","authors":["Zehao Huang","Naiyan Wang"],"abstract":"Deep convolutional neural networks have liberated its extraordinary power on\nvarious tasks. However, it is still very challenging to deploy state-of-the-art\nmodels into real-world applications due to their high computational complexity.\nHow can we design a compact and effective network without massive experiments\nand expert knowledge? In this paper, we propose a simple and effective\nframework to learn and prune deep models in an end-to-end manner. In our\nframework, a new type of parameter -- scaling factor is first introduced to\nscale the outputs of specific structures, such as neurons, groups or residual\nblocks. Then we add sparsity regularizations on these factors, and solve this\noptimization problem by a modified stochastic Accelerated Proximal Gradient\n(APG) method. By forcing some of the factors to zero, we can safely remove the\ncorresponding structures, thus prune the unimportant parts of a CNN. Comparing\nwith other structure selection methods that may need thousands of trials or\niterative fine-tuning, our method is trained fully end-to-end in one training\npass without bells and whistles. We evaluate our method, Sparse Structure\nSelection with several state-of-the-art CNNs, and demonstrate very promising\nresults with adaptive depth and width selection.","url_abs":"http://arxiv.org/abs/1707.01213v3","url_pdf":"http://arxiv.org/pdf/1707.01213v3.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":"data-driven-sparse-structure-selection-for","repo_url":"https://github.com/huangzehao/sparse-structure-selection","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"mxnet","reach":{"status":"unanswered"}},{"paper_slug":"data-driven-sparse-structure-selection-for","repo_url":"https://github.com/Pokemon-Huang/sparse-structure-selection-PyTorch","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":0,"framework":"pytorch","reach":{"status":"unanswered"}}],"tasks":[],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1707.01213","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}