{"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-bayesian-pruning-via-log-normal","title":"Structured Bayesian Pruning via Log-Normal Multiplicative Noise","arxiv_id":"1705.07283","date":"2017-05-20","proceeding":"NeurIPS 2017 12","authors":["Kirill Neklyudov","Dmitry Molchanov","Arsenii Ashukha","Dmitry Vetrov"],"abstract":"Dropout-based regularization methods can be regarded as injecting random\nnoise with pre-defined magnitude to different parts of the neural network\nduring training. It was recently shown that Bayesian dropout procedure not only\nimproves generalization but also leads to extremely sparse neural architectures\nby automatically setting the individual noise magnitude per weight. However,\nthis sparsity can hardly be used for acceleration since it is unstructured. In\nthe paper, we propose a new Bayesian model that takes into account the\ncomputational structure of neural networks and provides structured sparsity,\ne.g. removes neurons and/or convolutional channels in CNNs. To do this we\ninject noise to the neurons outputs while keeping the weights unregularized. We\nestablish the probabilistic model with a proper truncated log-uniform prior\nover the noise and truncated log-normal variational approximation that ensures\nthat the KL-term in the evidence lower bound is computed in closed-form. The\nmodel leads to structured sparsity by removing elements with a low SNR from the\ncomputation graph and provides significant acceleration on a number of deep\nneural architectures. The model is easy to implement as it can be formulated as\na separate dropout-like layer.","url_abs":"http://arxiv.org/abs/1705.07283v2","url_pdf":"http://arxiv.org/pdf/1705.07283v2.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-bayesian-pruning-via-log-normal","repo_url":"https://github.com/necludov/group-sparsity-sbp","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"tf","reach":{"status":"unanswered"}},{"paper_slug":"structured-bayesian-pruning-via-log-normal","repo_url":"https://github.com/ars-ashuha/variational-dropout-sparsifies-dnn","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"tf","reach":{"status":"ok","spdx":"GPL-3.0"}},{"paper_slug":"structured-bayesian-pruning-via-log-normal","repo_url":"https://github.com/maxblumental/variational-drouput","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"unanswered"}},{"paper_slug":"structured-bayesian-pruning-via-log-normal","repo_url":"https://github.com/senya-ashukha/variational-dropout-sparsifies-dnn","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"tf","reach":{"status":"ok","spdx":"GPL-3.0"}},{"paper_slug":"structured-bayesian-pruning-via-log-normal","repo_url":"https://github.com/xaosina/Struct-Sparse-Pruning","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"unanswered"}}],"tasks":[],"methods":[{"method_slug":"dropout","method_name":"Dropout"}],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1705.07283","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}