Sign inGet started
Paper · arXiv

On the Convergence of Adam, Revisited

We show that projected Adam for online optimization with arbitrary moment decay parameters $β_1,β_2\in[0,1)$ can have average regret bounded away from zero. A similar result of Reddi-Kale-Kumar from 2018 required $β_1<\sqrt{β_2}$. Similar to their result, we use a three-periodic sequence of linear functions on $[-1,1]$ with slopes $c,-1,-1$, though we use $c$ slightly larger than $2$. This nonzero average regret result extends to Adam variants such as AdamW, RMSProp, NAdam, Adan, AdaMax, Muon, and to an i.i.d. variant of the three-periodic sequence of slopes for Adam.

AuthorsSteven Heilman, Sampad Mohanty
Primary categorycs.LG
Published3 July 2026
Last updated at source3 July 2026
cs.LGmath.OCstat.ML
Embed this entry
<iframe src="https://silytics.si/embed/item/ad2af3a9-88f0-4ef4-b100-feb54f40d49a" width="420" height="190" style="border:0" loading="lazy" title="Silytics entry"></iframe>

Free, no tracking. More about feeds and embeds.

Silytics shows public metadata from arXiv and links to the original. Information may be incomplete or out of date. Last refreshed 17 October 2026. Licenses and terms of the original apply to the work itself. Not financial, legal or professional advice.