{"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/structured-pruning-of-deep-convolutional","title":"Structured Pruning of Deep Convolutional Neural Networks","arxiv_id":"1512.08571","date":"2015-12-29","proceeding":null,"authors":["Sajid Anwar","Kyuyeon Hwang","Wonyong Sung"],"abstract":"Real time application of deep learning algorithms is often hindered by high\ncomputational complexity and frequent memory accesses. Network pruning is a\npromising technique to solve this problem. However, pruning usually results in\nirregular network connections that not only demand extra representation efforts\nbut also do not fit well on parallel computation. We introduce structured\nsparsity at various scales for convolutional neural networks, which are channel\nwise, kernel wise and intra kernel strided sparsity. This structured sparsity\nis very advantageous for direct computational resource savings on embedded\ncomputers, parallel computing environments and hardware based systems. To\ndecide the importance of network connections and paths, the proposed method\nuses a particle filtering approach. The importance weight of each particle is\nassigned by computing the misclassification rate with corresponding\nconnectivity pattern. The pruned network is re-trained to compensate for the\nlosses due to pruning. While implementing convolutions as matrix products, we\nparticularly show that intra kernel strided sparsity with a simple constraint\ncan significantly reduce the size of kernel and feature map matrices. The\npruned network is finally fixed point optimized with reduced word length\nprecision. This results in significant reduction in the total storage size\nproviding advantages for on-chip memory based implementations of deep neural\nnetworks.","url_abs":"http://arxiv.org/abs/1512.08571v1","url_pdf":"http://arxiv.org/pdf/1512.08571v1.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":"structured-pruning-of-deep-convolutional","repo_url":"https://github.com/PaginDm/keras-L1-pruning","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"tf","reach":{"status":"ok","spdx":"MIT"}}],"tasks":[{"task_slug":"network-pruning","task_name":"Network Pruning"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"syntology_url":null,"atlas_url":"https://app.syntology.ai/?focus=1512.08571","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}