{"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/ssn-learning-sparse-switchable-normalization","title":"SSN: Learning Sparse Switchable Normalization via SparsestMax","arxiv_id":"1903.03793","date":"2019-03-09","proceeding":"CVPR 2019 6","authors":["Wenqi Shao","Tianjian Meng","Jingyu Li","Ruimao Zhang","Yudian Li","Xiaogang Wang","Ping Luo"],"abstract":"Normalization methods improve both optimization and generalization of\nConvNets. To further boost performance, the recently-proposed switchable\nnormalization (SN) provides a new perspective for deep learning: it learns to\nselect different normalizers for different convolution layers of a ConvNet.\nHowever, SN uses softmax function to learn importance ratios to combine\nnormalizers, leading to redundant computations compared to a single normalizer.\n  This work addresses this issue by presenting Sparse Switchable Normalization\n(SSN) where the importance ratios are constrained to be sparse. Unlike $\\ell_1$\nand $\\ell_0$ constraints that impose difficulties in optimization, we turn this\nconstrained optimization problem into feed-forward computation by proposing\nSparsestMax, which is a sparse version of softmax. SSN has several appealing\nproperties. (1) It inherits all benefits from SN such as applicability in\nvarious tasks and robustness to a wide range of batch sizes. (2) It is\nguaranteed to select only one normalizer for each normalization layer, avoiding\nredundant computations. (3) SSN can be transferred to various tasks in an\nend-to-end manner. Extensive experiments show that SSN outperforms its\ncounterparts on various challenging benchmarks such as ImageNet, Cityscapes,\nADE20K, and Kinetics.","url_abs":"http://arxiv.org/abs/1903.03793v1","url_pdf":"http://arxiv.org/pdf/1903.03793v1.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":"ssn-learning-sparse-switchable-normalization","repo_url":"https://github.com/switchablenorms/Sparse_SwitchNorm","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok"}}],"tasks":[],"methods":[{"method_slug":"batch-normalization","method_name":"Batch Normalization"},{"method_slug":"convolution","method_name":"Convolution"},{"method_slug":"instance-normalization","method_name":"Instance Normalization"},{"method_slug":"layer-normalization","method_name":"Layer Normalization"},{"method_slug":"softmax","method_name":"Softmax"},{"method_slug":"sparse-switchable-normalization","method_name":"Sparse Switchable Normalization"},{"method_slug":"sparsemax","method_name":"Sparsemax"}],"datasets_introduced":[],"methods_introduced":[{"slug":"sparse-switchable-normalization","name":"Sparse Switchable Normalization","full_name":"Sparse Switchable Normalization"}],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1903.03793","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}