{"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/efficient-sparse-winograd-convolutional","title":"Efficient Sparse-Winograd Convolutional Neural Networks","arxiv_id":"1802.06367","date":"2018-02-18","proceeding":"ICLR 2018 1","authors":["Xingyu Liu","Jeff Pool","Song Han","William J. Dally"],"abstract":"Convolutional Neural Networks (CNNs) are computationally intensive, which\nlimits their application on mobile devices. Their energy is dominated by the\nnumber of multiplies needed to perform the convolutions. Winograd's minimal\nfiltering algorithm (Lavin, 2015) and network pruning (Han et al., 2015) can\nreduce the operation count, but these two methods cannot be directly combined\n$-$ applying the Winograd transform fills in the sparsity in both the weights\nand the activations. We propose two modifications to Winograd-based CNNs to\nenable these methods to exploit sparsity. First, we move the ReLU operation\ninto the Winograd domain to increase the sparsity of the transformed\nactivations. Second, we prune the weights in the Winograd domain to exploit\nstatic weight sparsity. For models on CIFAR-10, CIFAR-100 and ImageNet\ndatasets, our method reduces the number of multiplications by $10.4\\times$,\n$6.8\\times$ and $10.8\\times$ respectively with loss of accuracy less than\n$0.1\\%$, outperforming previous baselines by $2.0\\times$-$3.0\\times$. We also\nshow that moving ReLU to the Winograd domain allows more aggressive pruning.","url_abs":"http://arxiv.org/abs/1802.06367v1","url_pdf":"http://arxiv.org/pdf/1802.06367v1.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":"efficient-sparse-winograd-convolutional","repo_url":"https://github.com/xingyul/Sparse-Winograd-CNN","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":0,"framework":"tf","reach":null}],"tasks":[{"task_slug":"network-pruning","task_name":"Network Pruning"}],"methods":[{"method_slug":"pruning","method_name":"Pruning"},{"method_slug":"relu","method_name":"ReLU"}],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1802.06367","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}