A Resolution of the SS--RS--GD Inequalities
Not on the map. Why?
Abstract (arXiv)
Yun, Sra, and Jadbabaie (COLT 2021, open question) conjectured the SS--RS--GD inequalities: for well-conditioned symmetric matrices A1,dots,An, the operators Wss, Wrs, and Wgd that encode the expected iterate of single-shuffle SGD, random-reshuffle SGD, and gradient descent on a quadratic finite sum should satisfy [ |Wss|≤ | Wrs|≤ |Wgd|. ] The conjecture is resolved, bullet SS-RS inequality fails. Already for n=3, K=2, and d=4, we exhibit explicit PSD matrices whose condition number is arbitrarily close to 1, yet |Wss|>|Wrs|. bullet RS-GD inequality holds. For every symmetric Ai with bigl(1-frac14n2+1bigr)Ipreceq Aipreceq I, one has |Wrs|≤|Wgd|. The proof was found via GPT-5.5 Pro extended prompted by the author.
Key facts. Cited by 1 paper in the map (data through 14 Oct 2026). Builds on 7 of its 11 references that are in the map. First posted on arXiv 16 Jun 2026.
Builds on
7 of 11 references.
References · 7
- 2020Random Reshuffling: Simple Analysis with Vast Improvementscited by 115
- 2018Random Shuffling Beats SGD after Finite Epochscited by 67
- 2019How Good is SGD with Random Shuffling?cited by 64
- 2019SGD without Replacement: Sharper Rates for General Smooth Convex Functionscited by 61
- 2020SGD with shuffling: optimal rates without component convexity and large epoch requirementscited by 58
- 2020Closing the convergence gap of SGD without replacementcited by 57
- 2020Recht-R'e Noncommutative Arithmetic-Geometric Mean Conjecture is Falsecited by 16
Built on by
Citing papers per year: 2026: 1.
1 paper.
- 2026When the noncommutative AM-GM inequality holdscited by 0
- 2026When the noncommutative AM-GM inequality holds4 Oct 2026