{"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/shakeout-a-new-approach-to-regularized-deep","title":"Shakeout: A New Approach to Regularized Deep Neural Network Training","arxiv_id":"1904.06593","date":"2019-04-13","proceeding":null,"authors":["Guoliang Kang","Jun Li","DaCheng Tao"],"abstract":"Recent years have witnessed the success of deep neural networks in dealing\nwith a plenty of practical problems. Dropout has played an essential role in\nmany successful deep neural networks, by inducing regularization in the model\ntraining. In this paper, we present a new regularized training approach:\nShakeout. Instead of randomly discarding units as Dropout does at the training\nstage, Shakeout randomly chooses to enhance or reverse each unit's contribution\nto the next layer. This minor modification of Dropout has the statistical\ntrait: the regularizer induced by Shakeout adaptively combines $L_0$, $L_1$ and\n$L_2$ regularization terms. Our classification experiments with representative\ndeep architectures on image datasets MNIST, CIFAR-10 and ImageNet show that\nShakeout deals with over-fitting effectively and outperforms Dropout. We\nempirically demonstrate that Shakeout leads to sparser weights under both\nunsupervised and supervised settings. Shakeout also leads to the grouping\neffect of the input units in a layer. Considering the weights in reflecting the\nimportance of connections, Shakeout is superior to Dropout, which is valuable\nfor the deep model compression. Moreover, we demonstrate that Shakeout can\neffectively reduce the instability of the training process of the deep\narchitecture.","url_abs":"http://arxiv.org/abs/1904.06593v1","url_pdf":"http://arxiv.org/pdf/1904.06593v1.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":"shakeout-a-new-approach-to-regularized-deep","repo_url":"https://github.com/kgl-prml/shakeout-for-caffe","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":0,"framework":"none","reach":null}],"tasks":[{"task_slug":"model-compression","task_name":"Model Compression"}],"methods":[{"method_slug":"dropout","method_name":"Dropout"}],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":null,"mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}