Convergence Analysis of STORM Under Different Geometries

Abstract

Stochastic recursive momentum (STORM) achieves fast convergence for nonconvex optimization via the variance reduction effect, but existing analyses rely on the strong average smoothness assumption. In this paper, we study the convergence of STORM for different objectives without average smoothness. We first revisit the results under average smoothness, obtaining the O(T^−1/3) bound for nonconvex objectives and the O(σ²/(μT)) bound for last-iterate output under the μ-Polyak--Łojasiewicz~(PL) condition. Without average smoothness, we design an auxiliary sequence and compare the STORM update with it in the analysis. With the help of this sequence, we prove that STORM still attains an O(T^−1/4) rate for nonconvex objectives, which is optimal under standard smoothness. For convex and λ-strongly convex objectives, we further prove averaged and last-iterate bounds with optimal rates of O(σR/√T) and O(σ²/(λT)), respectively. All the obtained results use the same STORM recursion with different hyperparameter choices.

A zero-inflated mixed-effects spatial point process for grouped storm loss data
Lisa Gao, Sébastien Jessup, Tianxing Yan
arXiv21 Sep 2026PreprintarXiv licenceFull textStatisticsarXiv
SurgeGen: A Hybrid Generative Diffusion Framework for Storm Surge Scenario Synthesis
Shunan Zheng, John J. Hasenbein
arXiv3 Sep 2026PreprintCC BY 4.0Full textDynamical systemsarXiv
When Protection Hurts: A Cytokine Storm Theory of Supply Chain Overreaction
Gurkirat Wadhwa
arXiv24 Oct 2026PreprintCC BY 4.0Full textOptimisation & controlarXiv
Singular Hodge theory for combinatorial geometries
Tom Braden, June Huh, Jacob P. Matherne +2
Journal of the American Mathematical Society1 Jul 2026Peer-reviewedLicence not statedFull text55 citedCombinatorics & discrete mathematicsOpenAlex
A Probabilistic Trajectory and Dispersion Study of Martian Tumbleweed Rover Swarms in Global Dust Storm Conditions
Emma-Catherine Belhadfa, Gabriele MT D'Eleuterio
arXiv29 Sep 2026PreprintCC BY 4.0Full textInstrumentation & methods Planetary science & exoplanetsarXiv

Cite

BibTeX RIS

BibTeX

@misc{jiang2026convergence,
  title = {{Convergence Analysis of STORM Under Different Geometries}},
  author = {Wei Jiang and Yibo Wang and Wenhao Yang and Rui Yan and Lijun Zhang and Zechao Li},
  howpublished = {arXiv},
  year = {2026},
  month = oct,
  doi = {10.48550/arxiv.2610.01599},
  eprint = {2610.01599},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.01599},
  note = {CC BY 4.0},
}

RIS

TY  - UNPB
TI  - Convergence Analysis of STORM Under Different Geometries
AU  - Wei Jiang
AU  - Yibo Wang
AU  - Wenhao Yang
AU  - Rui Yan
AU  - Lijun Zhang
AU  - Zechao Li
PY  - 2026
DA  - 2026/10/01
PB  - arXiv
DO  - 10.48550/arxiv.2610.01599
UR  - https://arxiv.org/abs/2610.01599
AB  - Stochastic recursive momentum (STORM) achieves fast convergence for nonconvex optimization via the variance reduction effect, but existing analyses rely on the strong average smoothness assumption. In this paper, we study the convergence of STORM for different objectives without average smoothness. We first revisit the results under average smoothness, obtaining the O(T^−1/3) bound for nonconvex objectives and the O(σ^2/(μT)) bound for last-iterate output under the μ-Polyak--Łojasiewicz~(PL) condition. Without average smoothness, we design an auxiliary sequence and compare the STORM update with it in the analysis. With the help of this sequence, we prove that STORM still attains an O(T^−1/4) rate for nonconvex objectives, which is optimal under standard smoothness. For convex and λ-strongly convex objectives, we further prove averaged and last-iterate bounds with optimal rates of O(σR/√T) and O(σ^2/(λT)), respectively. All the obtained results use the same STORM recursion with different hyperparameter choices.
ER  -

Abstract

Related

Cite