{"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/aggregated-momentum-stability-through-passive","title":"Aggregated Momentum: Stability Through Passive Damping","arxiv_id":"1804.00325","date":"2018-04-01","proceeding":"ICLR 2019 5","authors":["James Lucas","Shengyang Sun","Richard Zemel","Roger Grosse"],"abstract":"Momentum is a simple and widely used trick which allows gradient-based\noptimizers to pick up speed along low curvature directions. Its performance\ndepends crucially on a damping coefficient $\\beta$. Large $\\beta$ values can\npotentially deliver much larger speedups, but are prone to oscillations and\ninstability; hence one typically resorts to small values such as 0.5 or 0.9. We\npropose Aggregated Momentum (AggMo), a variant of momentum which combines\nmultiple velocity vectors with different $\\beta$ parameters. AggMo is trivial\nto implement, but significantly dampens oscillations, enabling it to remain\nstable even for aggressive $\\beta$ values such as 0.999. We reinterpret\nNesterov's accelerated gradient descent as a special case of AggMo and analyze\nrates of convergence for quadratic objectives. Empirically, we find that AggMo\nis a suitable drop-in replacement for other momentum methods, and frequently\ndelivers faster convergence.","url_abs":"http://arxiv.org/abs/1804.00325v3","url_pdf":"http://arxiv.org/pdf/1804.00325v3.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":"aggregated-momentum-stability-through-passive","repo_url":"https://github.com/AtheMathmo/AggMo","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok"}}],"tasks":[],"methods":[{"method_slug":"aggmo","method_name":"AggMo"},{"method_slug":"speed","method_name":"SPEED"}],"datasets_introduced":[],"methods_introduced":[{"slug":"aggmo","name":"AggMo","full_name":"AggMo"}],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1804.00325","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}