{"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/augment-your-batch-better-training-with","title":"Augment your batch: better training with larger batches","arxiv_id":"1901.09335","date":"2019-01-27","proceeding":null,"authors":["Elad Hoffer","Tal Ben-Nun","Itay Hubara","Niv Giladi","Torsten Hoefler","Daniel Soudry"],"abstract":"Large-batch SGD is important for scaling training of deep neural networks.\nHowever, without fine-tuning hyperparameter schedules, the generalization of\nthe model may be hampered. We propose to use batch augmentation: replicating\ninstances of samples within the same batch with different data augmentations.\nBatch augmentation acts as a regularizer and an accelerator, increasing both\ngeneralization and performance scaling. We analyze the effect of batch\naugmentation on gradient variance and show that it empirically improves\nconvergence for a wide variety of deep neural networks and datasets. Our\nresults show that batch augmentation reduces the number of necessary SGD\nupdates to achieve the same accuracy as the state-of-the-art. Overall, this\nsimple yet effective method enables faster training and better generalization\nby allowing more computational resources to be used concurrently.","url_abs":"http://arxiv.org/abs/1901.09335v1","url_pdf":"http://arxiv.org/pdf/1901.09335v1.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":"augment-your-batch-better-training-with","repo_url":"https://github.com/vaapopescu/gradient-pruning","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok","spdx":"MIT"}}],"tasks":[],"methods":[{"method_slug":"sgd","method_name":"SGD"}],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1901.09335","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}