pith. sign in

arxiv: 2607.01755 · v1 · pith:7KSFJSBCnew · submitted 2026-07-02 · 🧮 math.OC · cs.AI· cs.LG

Decentralized Stochastic Subgradient-type Methods with Communication Compression for Nonsmooth Nonconvex Optimization

Siyuan Zhang , Nachuan Xiao , Xin Liu This is my paper

Pith reviewed 2026-07-03 08:45 UTC · model grok-4.3

classification 🧮 math.OC cs.AIcs.LG
keywords decentralized optimizationcommunication compressionnonsmooth nonconvex optimizationsubgradient methodsdifferential inclusionsstochastic optimizationconsensus errorgradient tracking
0
0 comments X

The pith

Relating iterates to differential inclusion trajectories establishes global convergence of compressed decentralized subgradient methods for nonsmooth nonconvex problems.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper proposes a general framework unifying decentralized stochastic subgradient-type methods that use communication compression, either unbiased or contractive with error compensation. It connects the consensus-error iterates and averaged iterates to trajectories of continuous-time differential inclusions. This connection yields global convergence guarantees even when objective functions are nonsmooth and lack Clarke regularity. The framework also supports construction of new methods, including ones with sign-based regularization and gradient-tracking momentum. Preliminary experiments illustrate the resulting communication-accuracy trade-offs.

Core claim

By relating the consensus-error iterates and the averaged iterates to the trajectories of continuous-time differential inclusions, global convergence is established for all methods encompassed by the framework when the objective functions are nonsmooth and lack Clarke regularity.

What carries the argument

Relating consensus-error iterates and averaged iterates to trajectories of continuous-time differential inclusions

If this is right

  • Global convergence holds for every method inside the framework under the stated nonsmooth nonconvex conditions.
  • Both unbiased compression and contractive compression with error compensation are covered by the same convergence argument.
  • New methods built inside the framework, such as sign-based regularization variants and gradient-tracking momentum variants, inherit the global convergence guarantee.
  • Numerical experiments confirm practical viability and expose the communication-accuracy trade-off for the newly constructed methods.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The differential-inclusion technique may extend to deriving convergence rates or stability margins not addressed in the paper.
  • Similar continuous-time embeddings could analyze other distributed algorithms that combine compression with nonsmooth objectives.
  • In large-scale settings the framework suggests that communication budgets can be reduced while preserving global convergence for irregular loss landscapes.

Load-bearing premise

The discrete algorithm dynamics are accurately captured by the continuous-time differential inclusions for establishing global convergence.

What would settle it

A concrete instance of a nonsmooth nonconvex objective and a valid compression operator where an encompassed method diverges despite the differential inclusion trajectory predicting convergence.

Figures

Figures reproduced from arXiv: 2607.01755 by Nachuan Xiao, Siyuan Zhang, Xin Liu.

Figure 1
Figure 1. Figure 1: The comparison between CHOCO-SGD and CHOCO-SGD+ with the Random-10% opera [PITH_FULL_IMAGE:figures/full_fig_p023_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: The comparison between BEER and BEER+ with the rescaled [PITH_FULL_IMAGE:figures/full_fig_p023_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: The comparison between QSDGD and QSDGD+ with the [PITH_FULL_IMAGE:figures/full_fig_p024_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Numerical results of GD-based methods with contractive compression (Top-10% operator). [PITH_FULL_IMAGE:figures/full_fig_p024_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Numerical results of Gradient-tracking-based methods with contractive compression (Top [PITH_FULL_IMAGE:figures/full_fig_p025_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Numerical results of methods with unbiased compression ( [PITH_FULL_IMAGE:figures/full_fig_p025_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Numerical results of GD-based methods with contractive compression (rescaled [PITH_FULL_IMAGE:figures/full_fig_p025_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Numerical results of Gradient-tracking-based methods with contractive compression [PITH_FULL_IMAGE:figures/full_fig_p025_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Numerical results of GD-based methods with contractive compression (Random-10% oper [PITH_FULL_IMAGE:figures/full_fig_p026_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Numerical results of Gradient-tracking-based methods with contractive compression [PITH_FULL_IMAGE:figures/full_fig_p026_10.png] view at source ↗
read the original abstract

In this paper, we consider the nonsmooth nonconvex decentralized optimization problem, where inter-agent communication is compressed. We propose a general framework that unifies various decentralized stochastic subgradient-type methods with unbiased compression and contractive compression with error compensation. By relating the consensus-error iterates and the averaged iterates to the trajectories of continuous-time differential inclusions, we establish global convergence for all methods encompassed by our framework when the objective functions are nonsmooth and lack Clarke regularity. Based on our framework, we further develop several compression-based methods, including decentralized stochastic subgradient methods utilizing sign-based regularization and gradient-tracking momentum. Preliminary numerical experiments empirically support our theoretical results and highlight the communication-accuracy trade-off of the newly developed methods.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

1 major / 2 minor

Summary. The paper proposes a general framework unifying various decentralized stochastic subgradient-type methods that incorporate communication compression (covering both unbiased operators and contractive operators with error compensation) for nonsmooth nonconvex decentralized optimization problems. It claims to establish global convergence of all methods in the framework by relating the consensus-error iterates and the averaged iterates to trajectories of continuous-time differential inclusions, without requiring Clarke regularity of the objective functions. Based on the framework, new methods are developed (including sign-based regularization and gradient-tracking momentum variants), and preliminary numerical experiments are presented to support the theory and illustrate communication-accuracy trade-offs.

Significance. If the convergence result holds with the claimed generality, the work would be a meaningful contribution to decentralized optimization by extending convergence guarantees to nonsmooth nonconvex settings without Clarke regularity while accommodating practical compression schemes. The unification of compression types and the differential-inclusion technique for handling nonsmoothness are technical strengths; the new method variants and empirical illustrations of the trade-off add applied value for distributed learning scenarios.

major comments (1)
  1. [Abstract and convergence analysis] Abstract and convergence analysis section: the central claim that relating consensus-error and averaged iterates to DI trajectories yields global convergence for nonsmooth objectives lacking Clarke regularity is load-bearing. The discrete dynamics include stochastic subgradient noise plus compression-induced errors; it is unclear how the approximation error between the discrete stochastic compressed iterates and the DI trajectories is controlled uniformly as the step-size tends to zero, given that the DI may admit multiple solutions or lack a uniform Lyapunov decrease without regularity. Explicit error bounds or lemmas establishing this limit (independent of the nonsmoothness) are needed to secure the result.
minor comments (2)
  1. [Abstract] Abstract: the statement that 'preliminary numerical experiments empirically support our theoretical results' would be strengthened by including at least one quantitative performance metric or setup detail rather than remaining purely qualitative.
  2. [Framework section] Notation and assumptions: the precise definitions of the unbiased and contractive-with-error-compensation properties for the compression operators should be stated with equation numbers in the framework section for immediate reference when reading the convergence argument.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the careful reading and insightful comment on the convergence analysis. We address the concern below and will revise the manuscript to provide the requested explicit bounds.

read point-by-point responses

  1. Referee: [Abstract and convergence analysis] Abstract and convergence analysis section: the central claim that relating consensus-error and averaged iterates to DI trajectories yields global convergence for nonsmooth objectives lacking Clarke regularity is load-bearing. The discrete dynamics include stochastic subgradient noise plus compression-induced errors; it is unclear how the approximation error between the discrete stochastic compressed iterates and the DI trajectories is controlled uniformly as the step-size tends to zero, given that the DI may admit multiple solutions or lack a uniform Lyapunov decrease without regularity. Explicit error bounds or lemmas establishing this limit (independent of the nonsmoothness) are needed to secure the result.

    Authors: We appreciate this observation on the load-bearing step of the analysis. The manuscript adapts standard stochastic approximation arguments to show that the consensus-error and averaged iterates track DI trajectories in the limit, with the compression operators (unbiased or contractive with error compensation) and stochastic subgradient noise incorporated via bounded variance assumptions. However, we agree that the uniform control of the approximation error as the step-size tends to zero—particularly when the DI may have multiple solutions and without a uniform Lyapunov function—would benefit from an explicit lemma independent of Clarke regularity. In the revision we will insert a new lemma (in the convergence analysis section or appendix) deriving such bounds using only the Lipschitz continuity of the compression maps, bounded subgradient norms, and standard martingale convergence tools; this will make the passage to the limit fully rigorous without additional regularity. revision: yes

Circularity Check

0 steps flagged

No significant circularity; convergence via standard DI techniques is self-contained

full rationale

The abstract and description indicate the central step relates discrete consensus-error and averaged iterates to continuous-time differential inclusion trajectories to obtain global convergence for nonsmooth nonconvex problems. This is a standard external technique (not a self-definition, fitted prediction, or self-citation reduction). No quotes or equations in the provided material exhibit any of the enumerated circular patterns; the framework unifies existing methods and develops new ones without the result being forced by its own inputs. The analysis is therefore self-contained against external DI theory.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

Based solely on abstract; relies on standard mathematical techniques for differential inclusions and domain assumptions on compression operators. No free parameters, invented entities, or ad-hoc axioms explicitly introduced.

axioms (2)
  • domain assumption Compression operators are either unbiased or contractive with error compensation.
    Explicitly referenced as the types of compression unified in the framework.
  • domain assumption Differential inclusions accurately model the discrete consensus and averaged iterates for convergence analysis.
    Central to the global convergence proof for nonsmooth nonconvex cases.

pith-pipeline@v0.9.1-grok · 5651 in / 1101 out tokens · 31334 ms · 2026-07-03T08:45:53.666250+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

76 extracted references · 12 canonical work pages · 1 internal anchor

  1. [1]

    Qsgd: Communication-efficient sgd via gradient quantization and encoding.Advances in neural in- formation processing systems, 30, 2017

    Dan Alistarh, Demjan Grubic, Jerry Li, Ryota Tomioka, and Milan Vojnovic. Qsgd: Communication-efficient sgd via gradient quantization and encoding.Advances in neural in- formation processing systems, 30, 2017

    2017

  2. [2]

    The convergence of sparsified gradient methods.Advances in Neural Information Pro- cessing Systems, 31, 2018

    Dan Alistarh, Torsten Hoefler, Mikael Johansson, Nikola Konstantinov, Sarit Khirirat, and C´edric Renggli. The convergence of sparsified gradient methods.Advances in Neural Information Pro- cessing Systems, 31, 2018

    2018

  3. [3]

    Dynamics of stochastic approximation algorithms

    Michel Bena ¨ım. Dynamics of stochastic approximation algorithms. InSeminaire de Probabilites XXXIII, pages 1–68. Springer, 2006

    2006

  4. [4]

    Stochastic approximations and differential inclusions.SIAM Journal on Control and Optimization, 44(1):328–348, 2005

    Michel Bena ¨ım, Josef Hofbauer, and Sylvain Sorin. Stochastic approximations and differential inclusions.SIAM Journal on Control and Optimization, 44(1):328–348, 2005

    2005

  5. [5]

    Signsgd: Compressed optimisation for non-convex problems

    Jeremy Bernstein, Yu-Xiang Wang, Kamyar Azizzadenesheli, and Animashree Anandkumar. Signsgd: Compressed optimisation for non-convex problems. InInternational Conference on Ma- chine Learning, pages 560–569. PMLR, 2018

    2018

  6. [6]

    signSGD with Majority Vote is Communication Efficient And Fault Tolerant

    Jeremy Bernstein, Jiawei Zhao, Kamyar Azizzadenesheli, and Anima Anandkumar. signsgd with majority vote is communication efficient and fault tolerant.arXiv preprint arXiv:1810.05291, 2018

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  7. [7]

    On biased com- pression for distributed learning.Journal of Machine Learning Research, 24(276):1–50, 2023

    Aleksandr Beznosikov, Samuel Horv ´ath, Peter Richt ´arik, and Mher Safaryan. On biased com- pression for distributed learning.Journal of Machine Learning Research, 24(276):1–50, 2023

    2023

  8. [8]

    Subgradient-based neural networks for nonsmooth nonconvex optimization problems.IEEE Transactions on Neural Networks, 20(6):1024–1038, 2009

    Wei Bian and Xiaoping Xue. Subgradient-based neural networks for nonsmooth nonconvex optimization problems.IEEE Transactions on Neural Networks, 20(6):1024–1038, 2009

    2009

  9. [9]

    Clarke subgradients of strati- fiable functions.SIAM Journal on Optimization, 18(2):556–572, 2007

    J ´erˆome Bolte, Aris Daniilidis, Adrian Lewis, and Masahiro Shiota. Clarke subgradients of strati- fiable functions.SIAM Journal on Optimization, 18(2):556–572, 2007

    2007

  10. [10]

    A mathematical model for automatic differentiation in machine learning.Advances in Neural Information Processing Systems, 33:10809–10819, 2020

    J ´erˆome Bolte and Edouard Pauwels. A mathematical model for automatic differentiation in machine learning.Advances in Neural Information Processing Systems, 33:10809–10819, 2020

    2020

  11. [11]

    Conservative set valued fields, automatic differentiation, stochastic gradient methods and deep learning.Mathematical Programming, 188(1):19–51, 2021

    J ´erˆome Bolte and Edouard Pauwels. Conservative set valued fields, automatic differentiation, stochastic gradient methods and deep learning.Mathematical Programming, 188(1):19–51, 2021

    2021

  12. [12]

    Federated learning and privacy: Building privacy-preserving systems for machine learning and data science on decentralized data.Queue, 19(5):87–114, 2021

    Kallista Bonawitz, Peter Kairouz, Brendan McMahan, and Daniel Ramage. Federated learning and privacy: Building privacy-preserving systems for machine learning and data science on decentralized data.Queue, 19(5):87–114, 2021

    2021

  13. [13]

    Springer, 2009

    Vivek S Borkar.Stochastic approximation: a dynamical systems viewpoint, volume 48. Springer, 2009

    2009

  14. [14]

    Springer, 2008

    Vivek S Borkar and Vivek S Borkar.Stochastic approximation: a dynamical systems viewpoint, vol- ume 9. Springer, 2008

    2008

  15. [15]

    An inertial newton algo- rithm for deep learning.The Journal of Machine Learning Research, 22(1):5977–6007, 2021

    Camille Castera, J ´erˆome Bolte, C ´edric F´evotte, and Edouard Pauwels. An inertial newton algo- rithm for deep learning.The Journal of Machine Learning Research, 22(1):5977–6007, 2021

    2021

  16. [16]

    A decentralized framework for server- less edge computing in the internet of things.IEEE Transactions on Network and Service Manage- ment, 18(2):2166–2180, 2020

    Claudio Cicconetti, Marco Conti, and Andrea Passarella. A decentralized framework for server- less edge computing in the internet of things.IEEE Transactions on Network and Service Manage- ment, 18(2):2166–2180, 2020. 27

    2020

  17. [17]

    Springer Science & Business Media, 2008

    Francis H Clarke, Yuri S Ledyaev, Ronald J Stern, and Peter R Wolenski.Nonsmooth analysis and control theory, volume 178. Springer Science & Business Media, 2008

    2008

  18. [18]

    SIAM, 1990

    Frank H Clarke.Optimization and nonsmooth analysis, volume 5. SIAM, 1990

    1990

  19. [19]

    Stochastic subgradient method converges on tame functions.Foundations of Computational Mathematics, 20(1):119–154, 2020

    Damek Davis, Dmitriy Drusvyatskiy, Sham Kakade, and Jason D Lee. Stochastic subgradient method converges on tame functions.Foundations of Computational Mathematics, 20(1):119–154, 2020

    2020

  20. [20]

    Edge intelligence: The confluence of edge computing and artificial intelligence.IEEE Internet of Things Journal, 7(8):7457–7469, 2020

    Shuiguang Deng, Hailiang Zhao, Weijia Fang, Jianwei Yin, Schahram Dustdar, and Albert Y Zomaya. Edge intelligence: The confluence of edge computing and artificial intelligence.IEEE Internet of Things Journal, 7(8):7457–7469, 2020

    2020

  21. [21]

    Convergence rates of distributed gra- dient methods under random quantization: A stochastic approximation approach.IEEE Trans- actions on Automatic Control, 66(10):4469–4484, 2020

    Thinh T Doan, Siva Theja Maguluri, and Justin Romberg. Convergence rates of distributed gra- dient methods under random quantization: A stochastic approximation approach.IEEE Trans- actions on Automatic Control, 66(10):4469–4484, 2020

    2020

  22. [22]

    Stochastic methods for composite and weakly convex optimiza- tion problems.SIAM Journal on Optimization, 28(4):3229–3259, 2018

    John C Duchi and Feng Ruan. Stochastic methods for composite and weakly convex optimiza- tion problems.SIAM Journal on Optimization, 28(4):3229–3259, 2018

    2018

  23. [23]

    Ef21 with bells & whistles: Practical algorithmic extensions of modern error feedback.arXiv preprint arXiv:2110.03294, 2021

    Ilyas Fatkhullin, Igor Sokolov, Eduard Gorbunov, Zhize Li, and Peter Richt ´arik. Ef21 with bells & whistles: Practical algorithmic extensions of modern error feedback.arXiv preprint arXiv:2110.03294, 2021

    work page arXiv 2021

  24. [24]

    Consensus multi-agent reinforcement learning for volt-var control in power distribution networks.IEEE Transactions on Smart Grid, 12(4):3594– 3604, 2021

    Yuanqi Gao, Wei Wang, and Nanpeng Yu. Consensus multi-agent reinforcement learning for volt-var control in power distribution networks.IEEE Transactions on Smart Grid, 12(4):3594– 3604, 2021

    2021

  25. [25]

    Natural compression for distributed deep learning

    Samuel Horv ´oth, Chen-Yu Ho, Ludovit Horvath, Atal Narayan Sahu, Marco Canini, and Peter Richt´arik. Natural compression for distributed deep learning. InMathematical and Scientific Machine Learning, pages 129–141. PMLR, 2022

    2022

  26. [26]

    Lower bounds and nearly optimal algorithms in distributed learning with communication compression.Advances in Neural Infor- mation Processing Systems, 35:18955–18969, 2022

    Xinmeng Huang, Yiming Chen, Wotao Yin, and Kun Yuan. Lower bounds and nearly optimal algorithms in distributed learning with communication compression.Advances in Neural Infor- mation Processing Systems, 35:18955–18969, 2022

    2022

  27. [27]

    S-near-dgd: A flexible distributed stochastic gradient method for inexact communication.IEEE Transactions on Automatic Control, 68(2):1281–1287, 2022

    Charikleia Iakovidou and Ermin Wei. S-near-dgd: A flexible distributed stochastic gradient method for inexact communication.IEEE Transactions on Automatic Control, 68(2):1281–1287, 2022

    2022

  28. [28]

    Rustem Islamov, Yuan Gao, and Sebastian U. Stich. Near optimal decentralized optimization with compression and momentum tracking.CoRR, abs/2405.20114, 2024

    work page arXiv 2024

  29. [29]

    Lyapunov stability of the subgradient method with constant step size.Mathematical Programming, 202(1):387–396, 2023

    C ´edric Josz and Lexiao Lai. Lyapunov stability of the subgradient method with constant step size.Mathematical Programming, 202(1):387–396, 2023

    2023

  30. [30]

    Privacy-preserving decentralized learning framework for healthcare system.ACM Transactions on Multimedia Computing, Communications, and Applications (TOMM), 17(2s):1–24, 2021

    Harsh Kasyap and Somanath Tripathy. Privacy-preserving decentralized learning framework for healthcare system.ACM Transactions on Multimedia Computing, Communications, and Applications (TOMM), 17(2s):1–24, 2021

    2021

  31. [31]

    Decentralized deep learning with arbitrary communication compression.arXiv preprint arXiv:1907.09356, 2019

    Anastasia Koloskova, Tao Lin, Sebastian U Stich, and Martin Jaggi. Decentralized deep learning with arbitrary communication compression.arXiv preprint arXiv:1907.09356, 2019

    work page arXiv 1907

  32. [32]

    Decentralized stochastic optimization and gossip algorithms with compressed communication

    Anastasia Koloskova, Sebastian Stich, and Martin Jaggi. Decentralized stochastic optimization and gossip algorithms with compressed communication. InInternational Conference on Machine Learning, pages 3478–3487. PMLR, 2019

    2019

  33. [33]

    Nonsmooth nonconvex stochastic heavy ball.Journal of Optimization Theory and Applica- tions, 201(2):699–719, 2024

    Tam Le. Nonsmooth nonconvex stochastic heavy ball.Journal of Optimization Theory and Applica- tions, 201(2):699–719, 2024. 28

    2024

  34. [34]

    Xin Liu Lei Wang. A variance-reduced stochastic gradient tracking algorithm for decentralized optimization with orthogonality constraints.Journal of Industrial and Management Optimization, 19(10):7753–7776, 2023

    2023

  35. [35]

    Distributed subgradient method for multi- agent optimization with quantized communication.Mathematical Methods in the Applied Sciences, 40(4):1201–1213, 2017

    Jueyou Li, Guo Chen, Zhiyou Wu, and Xing He. Distributed subgradient method for multi- agent optimization with quantized communication.Mathematical Methods in the Applied Sciences, 40(4):1201–1213, 2017

    2017

  36. [36]

    Decentralized composite optimiza- tion with compression.arXiv preprint arXiv:2108.04448, 2021

    Yao Li, Xiaorui Liu, Jiliang Tang, Ming Yan, and Kun Yuan. Decentralized composite optimiza- tion with compression.arXiv preprint arXiv:2108.04448, 2021

    work page arXiv 2021

  37. [37]

    Xiangru Lian, Ce Zhang, Huan Zhang, Cho-Jui Hsieh, Wei Zhang, and Ji Liu. Can decentralized algorithms outperform centralized algorithms? a case study for decentralized parallel stochastic gradient descent.Advances in neural information processing systems, 30, 2017

    2017

  38. [38]

    A compressed gradient tracking method for decentralized optimization with linear convergence.IEEE Transactions on Automatic Control, 67(10):5622–5629, 2022

    Yiwei Liao, Zhuorui Li, Kun Huang, and Shi Pu. A compressed gradient tracking method for decentralized optimization with linear convergence.IEEE Transactions on Automatic Control, 67(10):5622–5629, 2022

    2022

  39. [39]

    A robust compressed push-pull method for decentralized nonconvex optimization.arXiv preprint arXiv:2408.01727, 2024

    Yiwei Liao, Zhuorui Li, Shi Pu, and Tsung-Hui Chang. A robust compressed push-pull method for decentralized nonconvex optimization.arXiv preprint arXiv:2408.01727, 2024

    work page arXiv 2024

  40. [40]

    Slota, Naigang Wang, Jie Chen, and Yangyang Xu

    Wei Liu, Anweshit Panda, Ujwal Pandey, Christopher Brissette, Yikang Shen, George M. Slota, Naigang Wang, Jie Chen, and Yangyang Xu. Compressed decentralized momentum stochastic gradient methods for nonconvex optimization, 2025

    2025

  41. [41]

    Learning- driven decentralized machine learning in resource-constrained wireless edge computing

    Zeyu Meng, Hongli Xu, Min Chen, Yang Xu, Yangming Zhao, and Chunming Qiao. Learning- driven decentralized machine learning in resource-constrained wireless edge computing. In IEEE INFOCOM 2021-IEEE Conference on Computer Communications, pages 1–10. IEEE, 2021

    2021

  42. [42]

    Critical random graphs: diameter and mixing time

    Asaf Nachmias and Yuval Peres. Critical random graphs: diameter and mixing time. 2008

    2008

  43. [43]

    Differential error feedback for communication-efficient decentralized learning.IEEE Transactions on Signal Processing, 2025

    Roula Nassif, Stefan Vlaski, Marco Carpentiero, Vincenzo Matta, and Ali H Sayed. Differential error feedback for communication-efficient decentralized learning.IEEE Transactions on Signal Processing, 2025

    2025

  44. [44]

    Network topology and communication- computation tradeoffs in decentralized optimization.Proceedings of the IEEE, 106(5):953–976, 2018

    Angelia Nedi ´c, Alex Olshevsky, and Michael G Rabbat. Network topology and communication- computation tradeoffs in decentralized optimization.Proceedings of the IEEE, 106(5):953–976, 2018

    2018

  45. [45]

    A method of solving a convex programming problem with con- vergence rate o\bigl(kˆ2\bigr)

    Yurii Evgen’evich Nesterov. A method of solving a convex programming problem with con- vergence rate o\bigl(kˆ2\bigr). InDoklady Akademii Nauk, volume 269, pages 543–547. Russian Academy of Sciences, 1983

    1983

  46. [46]

    Incremental without replacement sampling in nonconvex optimization.Jour- nal of Optimization Theory and Applications, 190(1):274–299, 2021

    Edouard Pauwels. Incremental without replacement sampling in nonconvex optimization.Jour- nal of Optimization Theory and Applications, 190(1):274–299, 2021

    2021

  47. [47]

    The perron-frobenius theorem: some of its applications.IEEE Signal Processing Magazine, 22(2):62–75, 2005

    S Unnikrishna Pillai, Torsten Suel, and Seunghun Cha. The perron-frobenius theorem: some of its applications.IEEE Signal Processing Magazine, 22(2):62–75, 2005

    2005

  48. [48]

    Some methods of speeding up the convergence of iteration methods.Ussr computational mathematics and mathematical physics, 4(5):1–17, 1964

    Boris T Polyak. Some methods of speeding up the convergence of iteration methods.Ussr computational mathematics and mathematical physics, 4(5):1–17, 1964

    1964

  49. [49]

    Noise reduction by swarming in social foraging.IEEE Transactions on Automatic Control, 61(12):4007–4013, 2016

    Shi Pu, Alfredo Garcia, and Zongli Lin. Noise reduction by swarming in social foraging.IEEE Transactions on Automatic Control, 61(12):4007–4013, 2016

    2016

  50. [50]

    Error compensated distributed sgd can be acceler- ated.Advances in Neural Information Processing Systems, 34:30401–30413, 2021

    Xun Qian, Peter Richt ´arik, and Tong Zhang. Error compensated distributed sgd can be acceler- ated.Advances in Neural Information Processing Systems, 34:30401–30413, 2021. 29

    2021

  51. [51]

    An ex- act quantized decentralized gradient descent algorithm.IEEE Transactions on Signal Processing, 67(19):4934–4947, 2019

    Amirhossein Reisizadeh, Aryan Mokhtari, Hamed Hassani, and Ramtin Pedarsani. An ex- act quantized decentralized gradient descent algorithm.IEEE Transactions on Signal Processing, 67(19):4934–4947, 2019

    2019

  52. [52]

    Ef21: A new, simpler, theoretically better, and practically faster error feedback.Advances in Neural Information Processing Systems, 34:4384–4396, 2021

    Peter Richt ´arik, Igor Sokolov, and Ilyas Fatkhullin. Ef21: A new, simpler, theoretically better, and practically faster error feedback.Advances in Neural Information Processing Systems, 34:4384–4396, 2021

    2021

  53. [53]

    Robbins and D

    H. Robbins and D. Siegmund. A convergence theorem for non negative almost supermartin- gales and some applications**research supported by nih grant 5-r01-gm-16895-03 and onr grant n00014-67-a-0108-0018. In Jagdish S. Rustagi, editor,Optimizing Methods in Statistics, pages 233–

  54. [54]

    Academic Press, 1971

    1971

  55. [55]

    A stochastic gradient tracking algorithm for decentral- ized optimization with inexact communication.arXiv preprint arXiv:2307.14942, 2023

    Suhail M Shah and Raghu Bollapragada. A stochastic gradient tracking algorithm for decentral- ized optimization with inexact communication.arXiv preprint arXiv:2307.14942, 2023

    work page arXiv 2023

  56. [56]

    Extra: An exact first-order algorithm for decen- tralized consensus optimization.SIAM Journal on Optimization, 25(2):944–966, 2015

    Wei Shi, Qing Ling, Gang Wu, and Wotao Yin. Extra: An exact first-order algorithm for decen- tralized consensus optimization.SIAM Journal on Optimization, 25(2):944–966, 2015

    2015

  57. [57]

    Squarm-sgd: Communication- efficient momentum sgd for decentralized optimization.IEEE Journal on Selected Areas in Infor- mation Theory, 2(3):954–969, 2021

    Navjot Singh, Deepesh Data, Jemin George, and Suhas Diggavi. Squarm-sgd: Communication- efficient momentum sgd for decentralized optimization.IEEE Journal on Selected Areas in Infor- mation Theory, 2(3):954–969, 2021

    2021

  58. [58]

    Sparq-sgd: Event-triggered and compressed communication in decentralized optimization.IEEE Transactions on Automatic Control, 68(2):721–736, 2022

    Navjot Singh, Deepesh Data, Jemin George, and Suhas Diggavi. Sparq-sgd: Event-triggered and compressed communication in decentralized optimization.IEEE Transactions on Automatic Control, 68(2):721–736, 2022

    2022

  59. [59]

    Sparsified sgd with memory

    Sebastian U Stich, Jean-Baptiste Cordonnier, and Martin Jaggi. Sparsified sgd with memory. Advances in neural information processing systems, 31, 2018

    2018

  60. [60]

    Distributed anonymous mobile robots: Formation of geometric patterns.SIAM Journal on Computing, 28(4):1347–1363, 1999

    Ichiro Suzuki and Masafumi Yamashita. Distributed anonymous mobile robots: Formation of geometric patterns.SIAM Journal on Computing, 28(4):1347–1363, 1999

    1999

  61. [61]

    Communication compression for decentralized training.Advances in Neural Information Processing Systems, 31, 2018

    Hanlin Tang, Shaoduo Gan, Ce Zhang, Tong Zhang, and Ji Liu. Communication compression for decentralized training.Advances in Neural Information Processing Systems, 31, 2018

    2018

  62. [62]

    Learning structured sparsity in deep neural networks.Advances in neural information processing systems, 29, 2016

    Wei Wen, Chunpeng Wu, Yandan Wang, Yiran Chen, and Hai Li. Learning structured sparsity in deep neural networks.Advances in neural information processing systems, 29, 2016

    2016

  63. [63]

    Terngrad: Ternary gradients to reduce communication in distributed deep learning.Advances in neural information processing systems, 30, 2017

    Wei Wen, Cong Xu, Feng Yan, Chunpeng Wu, Yandan Wang, Yiran Chen, and Hai Li. Terngrad: Ternary gradients to reduce communication in distributed deep learning.Advances in neural information processing systems, 30, 2017

    2017

  64. [64]

    Fast linear iterations for distributed averaging.Systems & Control Letters, 53(1):65–78, 2004

    Lin Xiao and Stephen Boyd. Fast linear iterations for distributed averaging.Systems & Control Letters, 53(1):65–78, 2004

    2004

  65. [65]

    Distributed average consensus with time-varying metropolis weights.Automatica, 1, 2006

    Lin Xiao, Stephen Boyd, and Sanjay Lall. Distributed average consensus with time-varying metropolis weights.Automatica, 1, 2006

    2006

  66. [66]

    Adam-family methods for nonsmooth optimization with convergence guarantees.arXiv preprint arXiv:2305.03938, 2023

    Nachuan Xiao, Xiaoyin Hu, Xin Liu, and Kim-Chuan Toh. Adam-family methods for nonsmooth optimization with convergence guarantees.arXiv preprint arXiv:2305.03938, 2023

    work page arXiv 2023

  67. [67]

    Convergence guarantees for stochastic sub- gradient methods in nonsmooth nonconvex optimization.arXiv preprint arXiv:2307.10053, 2023

    Nachuan Xiao, Xiaoyin Hu, and Kim-Chuan Toh. Convergence guarantees for stochastic sub- gradient methods in nonsmooth nonconvex optimization.arXiv preprint arXiv:2307.10053, 2023

    work page arXiv 2023

  68. [68]

    Compressed decentralized proximal stochastic gradient method for nonconvex composite problems with het- erogeneous data

    Yonggui Yan, Jie Chen, Pin-Yu Chen, Xiaodong Cui, Songtao Lu, and Yangyang Xu. Compressed decentralized proximal stochastic gradient method for nonconvex composite problems with het- erogeneous data. InInternational Conference on Machine Learning, pages 39035–39061. PMLR, 2023

    2023

  69. [69]

    Docom: Compressed decentralized optimization with near- optimal sample complexity.arXiv preprint arXiv:2202.00255, 2022

    Chung-Yiu Yau and Hoi-To Wai. Docom: Compressed decentralized optimization with near- optimal sample complexity.arXiv preprint arXiv:2202.00255, 2022. 30

    work page arXiv 2022

  70. [70]

    Zipml: Training linear models with end-to-end low precision, and a little bit of deep learning

    Hantian Zhang, Jerry Li, Kaan Kara, Dan Alistarh, Ji Liu, and Ce Zhang. Zipml: Training linear models with end-to-end low precision, and a little bit of deep learning. InInternational Conference on Machine Learning, pages 4035–4043. PMLR, 2017

    2017

  71. [71]

    Multi-agent reinforcement learning: A selec- tive overview of theories and algorithms.Handbook of reinforcement learning and control, pages 321–384, 2021

    Kaiqing Zhang, Zhuoran Yang, and Tamer Bas ¸ar. Multi-agent reinforcement learning: A selec- tive overview of theories and algorithms.Handbook of reinforcement learning and control, pages 321–384, 2021

    2021

  72. [72]

    Decentralized stochastic subgradient methods for nonsmooth nonconvex optimization.arXiv preprint arXiv:2403.11565, 2024

    Siyuan Zhang, Nachuan Xiao, and Xin Liu. Decentralized stochastic subgradient methods for nonsmooth nonconvex optimization.arXiv preprint arXiv:2403.11565, 2024

    work page arXiv 2024

  73. [73]

    Beer: Fasto(1/t)rate for decentralized nonconvex optimization with communication compression.Advances in Neural Information Processing Systems, 35:31653–31667, 2022

    Haoyu Zhao, Boyue Li, Zhize Li, Peter Richt ´arik, and Yuejie Chi. Beer: Fasto(1/t)rate for decentralized nonconvex optimization with communication compression.Advances in Neural Information Processing Systems, 35:31653–31667, 2022

    2022

  74. [74]

    Communication-efficient distributed blockwise momentum sgd with error-feedback.Advances in Neural Information Processing Systems, 32, 2019

    Shuai Zheng, Ziyue Huang, and James Kwok. Communication-efficient distributed blockwise momentum sgd with error-feedback.Advances in Neural Information Processing Systems, 32, 2019

    2019

  75. [75]

    Dp-csgp: Differentially private stochastic gradient push with compressed communication.arXiv preprint arXiv:2512.13583, 2025

    Zehan Zhu, Heng Zhao, Yan Huang, Joey Tianyi Zhou, Shouling Ji, and Jinming Xu. Dp-csgp: Differentially private stochastic gradient push with compressed communication.arXiv preprint arXiv:2512.13583, 2025

    work page arXiv 2025

  76. [76]

    Ensembles semi-analytiques

    Stanisław Łojasiewicz. Ensembles semi-analytiques. 1965. Appendix 6.1 Proof of Lemma 3.8 Proof.A straightforward calculation shows that, for anyN∈N ∗, aγk + k ∑ i=N k ∏ j=i (1−aγ j)aγi−1 + k ∏ j=N−1 (1−aγ j) =aγ k + k ∑ i=N+1 k ∏ j=i (1−aγ j)aγi−1 + k ∏ j=N (1−aγ j)aγ N−1 + k ∏ j=N (1−aγ j)(1−aγ N−1 ) =aγ k + k ∑ i=N+1 k ∏ j=i (1−aγ j)aγi−1 + k ∏ j=N (1−a...

    1965