{"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/accumulated-gradient-normalization","title":"Accumulated Gradient Normalization","arxiv_id":"1710.02368","date":"2017-10-06","proceeding":null,"authors":["Joeri Hermans","Gerasimos Spanakis","Rico Möckel"],"abstract":"This work addresses the instability in asynchronous data parallel\noptimization. It does so by introducing a novel distributed optimizer which is\nable to efficiently optimize a centralized model under communication\nconstraints. The optimizer achieves this by pushing a normalized sequence of\nfirst-order gradients to a parameter server. This implies that the magnitude of\na worker delta is smaller compared to an accumulated gradient, and provides a\nbetter direction towards a minimum compared to first-order gradients, which in\nturn also forces possible implicit momentum fluctuations to be more aligned\nsince we make the assumption that all workers contribute towards a single\nminima. As a result, our approach mitigates the parameter staleness problem\nmore effectively since staleness in asynchrony induces (implicit) momentum, and\nachieves a better convergence rate compared to other optimizers such as\nasynchronous EASGD and DynSGD, which we show empirically.","url_abs":"http://arxiv.org/abs/1710.02368v1","url_pdf":"http://arxiv.org/pdf/1710.02368v1.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":"accumulated-gradient-normalization","repo_url":"https://github.com/tmulc18/Distributed-TensorFlow-Guide","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"tf","reach":{"status":"ok","spdx":"MIT"}}],"tasks":[],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1710.02368","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}