{"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/accelerating-training-of-deep-neural-networks","title":"Accelerating Training of Deep Neural Networks with a Standardization Loss","arxiv_id":"1903.00925","date":"2019-03-03","proceeding":null,"authors":["Jasmine Collins","Johannes Balle","Jonathon Shlens"],"abstract":"A significant advance in accelerating neural network training has been the\ndevelopment of normalization methods, permitting the training of deep models\nboth faster and with better accuracy. These advances come with practical\nchallenges: for instance, batch normalization ties the prediction of individual\nexamples with other examples within a batch, resulting in a network that is\nheavily dependent on batch size. Layer normalization and group normalization\nare data-dependent and thus must be continually used, even at test-time. To\naddress the issues that arise from using explicit normalization techniques, we\npropose to replace existing normalization methods with a simple, secondary\nobjective loss that we term a standardization loss. This formulation is\nflexible and robust across different batch sizes and surprisingly, this\nsecondary objective accelerates learning on the primary training objective.\nBecause it is a training loss, it is simply removed at test-time, and no\nfurther effort is needed to maintain normalized activations. We find that a\nstandardization loss accelerates training on both small- and large-scale image\nclassification experiments, works with a variety of architectures, and is\nlargely robust to training across different batch sizes.","url_abs":"http://arxiv.org/abs/1903.00925v1","url_pdf":"http://arxiv.org/pdf/1903.00925v1.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":"accelerating-training-of-deep-neural-networks","repo_url":"https://github.com/lessw2020/auto-adaptive-ai","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok"}}],"tasks":[{"task_slug":"image-classification","task_name":"Image Classification"},{"task_slug":"image-classification","task_name":"image-classification"}],"methods":[{"method_slug":"batch-normalization","method_name":"Batch Normalization"},{"method_slug":"layer-normalization","method_name":"Layer Normalization"}],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":null,"mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}