{"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/exploiting-kernel-sparsity-and-entropy-for","title":"Exploiting Kernel Sparsity and Entropy for Interpretable CNN Compression","arxiv_id":"1812.04368","date":"2018-12-11","proceeding":"CVPR 2019 6","authors":["Yuchao Li","Shaohui Lin","Baochang Zhang","Jianzhuang Liu","David Doermann","Yongjian Wu","Feiyue Huang","Rongrong Ji"],"abstract":"Compressing convolutional neural networks (CNNs) has received ever-increasing\nresearch focus. However, most existing CNN compression methods do not interpret\ntheir inherent structures to distinguish the implicit redundancy. In this\npaper, we investigate the problem of CNN compression from a novel interpretable\nperspective. The relationship between the input feature maps and 2D kernels is\nrevealed in a theoretical framework, based on which a kernel sparsity and\nentropy (KSE) indicator is proposed to quantitate the feature map importance in\na feature-agnostic manner to guide model compression. Kernel clustering is\nfurther conducted based on the KSE indicator to accomplish high-precision CNN\ncompression. KSE is capable of simultaneously compressing each layer in an\nefficient way, which is significantly faster compared to previous data-driven\nfeature map pruning methods. We comprehensively evaluate the compression and\nspeedup of the proposed method on CIFAR-10, SVHN and ImageNet 2012. Our method\ndemonstrates superior performance gains over previous ones. In particular, it\nachieves 4.7 \\times FLOPs reduction and 2.9 \\times compression on ResNet-50\nwith only a Top-5 accuracy drop of 0.35% on ImageNet 2012, which significantly\noutperforms state-of-the-art methods.","url_abs":"http://arxiv.org/abs/1812.04368v2","url_pdf":"http://arxiv.org/pdf/1812.04368v2.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":"exploiting-kernel-sparsity-and-entropy-for","repo_url":"https://github.com/yuchaoli/KSE","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":0,"framework":"pytorch","reach":{"status":"gone","observed_at":"2026-09-18","how":"tree_404+repo_404"}}],"tasks":[{"task_slug":"clustering","task_name":"Clustering"},{"task_slug":"model-compression","task_name":"Model Compression"}],"methods":[{"method_slug":"pruning","method_name":"Pruning"}],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1812.04368","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}