pith. sign in

arxiv: 2604.21381 · v1 · submitted 2026-04-23 · 📡 eess.SY · cs.SY

Privacy-Preserving Distributed Stochastic Optimization with Homomorphic Encryption and Heterogeneous Stepsizes

Haoqiang Zhou , Chi Chen , Yongfeng Zhi , Huan Gao This is my paper

Pith reviewed 2026-05-09 21:06 UTC · model grok-4.3

classification 📡 eess.SY cs.SY
keywords distributed stochastic optimizationhomomorphic encryptionprivacy preservationPaillier cryptosystemheterogeneous stepsizesalmost sure convergencequantization error mitigation
0
0 comments X

The pith

A distributed stochastic gradient algorithm using Paillier encryption and heterogeneous random stepsizes shields data from curious agents and eavesdroppers while converging almost surely to the optimum.

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

The paper develops a method for multiple agents to jointly minimize a cost function by exchanging gradient information without exposing their private data. It encrypts the exchanged values with Paillier homomorphic encryption so that honest-but-curious neighbors and outside listeners cannot recover individual contributions. An attenuation factor is multiplied into the updates to keep the rounding errors introduced by encryption from accumulating. Heterogeneous time-varying random stepsizes are retained so that the iterates reach the global minimum with probability one. The result matters for sensor networks and collaborative learning where both accuracy and data confidentiality are required without a central trusted coordinator.

Core claim

The algorithm integrates Paillier homomorphic encryption with heterogeneous and time-varying random stepsizes and adds an attenuation factor to control quantization error, thereby delivering inherent privacy against internal honest-but-curious agents and external eavesdroppers without any trusted neighbors while guaranteeing almost-sure convergence to the optimal solution.

What carries the argument

The central mechanism is the combination of Paillier homomorphic encryption for secure sharing of gradients with an attenuation factor that damps quantization noise and heterogeneous time-varying random stepsizes that drive almost-sure convergence.

If this is right

  • Agents can share only encrypted values and still reach the same optimum as the unencrypted version.
  • Privacy holds against both internal participants who follow the protocol and external parties who intercept messages.
  • No trusted neighbor or third party is required for the privacy guarantee.
  • The method applies directly to distributed learning and sensor-network tasks that already use stochastic gradient updates.
  • Numerical tests confirm that the added encryption and attenuation do not destroy the convergence rate achieved by the random stepsizes.

Where Pith is reading between the lines

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

  • The same attenuation technique could be tested with other additively homomorphic schemes if their quantization noise behaves similarly.
  • In bandwidth-limited networks the encrypted messages might be compressed further without losing the convergence guarantee.
  • The random stepsizes already provide some noise; combining them with encryption noise could be analyzed to see whether differential privacy guarantees emerge for free.

Load-bearing premise

An attenuation factor exists that reduces encryption-induced quantization error enough to leave the almost-sure convergence property of the heterogeneous random stepsizes intact.

What would settle it

A concrete counterexample in which, for every choice of attenuation factor, either private local data can be recovered from the encrypted messages or the iterates fail to converge almost surely to the optimum.

Figures

Figures reproduced from arXiv: 2604.21381 by Chi Chen, Haoqiang Zhou, Huan Gao, Yongfeng Zhi.

Figure 1
Figure 1. Figure 1: Paillier-based secure interaction protocol. [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Undirected topology of the network. noise νij followed a uniform distribution on [−0.5, 0.5]. In our proposed algorithm, we set the quantization precision δ = 0.1 and the attenuation factor γ k = 1 1+0.1k0.81 . The stepsize matrix for agent i at iteration k was set as Λ k i = diag{λ k i1 , . . . , λk id}, where each diagonal entry was generated as λ k il = 0.005 k0.6 [PITH_FULL_IMAGE:figures/full_fig_p007… view at source ↗
Figure 3
Figure 3. Figure 3: Comparison of our algorithm with conventional [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
read the original abstract

Distributed stochastic optimization enables multi-agent collaboration in applications such as distributed learning and sensor networks, but also raises critical privacy concerns due to the involvement of sensitive data. While existing privacy-preserving approaches often face limitations in balancing accuracy with efficiency, we propose a novel distributed stochastic gradient descent algorithm that integrates Paillier homomorphic encryption with heterogeneous and time-varying random stepsizes. The proposed algorithm provides inherent privacy protection against both internal honest-but-curious agents and external eavesdroppers, without relying on any trusted neighbors. Furthermore, we incorporate an attenuation factor to effectively mitigate quantization error induced by the encryption process, ensuring almost sure convergence to the optimal solution while maintaining privacy preservation. Numerical simulations demonstrate the effectiveness and efficiency of the proposed approach.

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 distributed stochastic gradient descent algorithm integrating Paillier homomorphic encryption for privacy preservation with heterogeneous time-varying random stepsizes and an attenuation factor to mitigate quantization errors from encryption. It claims inherent privacy against internal honest-but-curious agents and external eavesdroppers without trusted neighbors, almost-sure convergence to the optimum, and demonstrates effectiveness via numerical simulations.

Significance. If the convergence analysis holds, the result would be significant for privacy-preserving distributed optimization, as it combines cryptographic tools with stochastic approximation techniques using heterogeneous stepsizes. The explicit handling of quantization via an attenuation factor while claiming a.s. convergence is a potential strength if the conditions are rigorously verified.

major comments (1)
  1. [Convergence analysis section / Theorem on a.s. convergence] Convergence analysis (likely §4 or Theorem on a.s. convergence): The claim that the attenuation factor mitigates quantization error while preserving almost-sure convergence relies on the heterogeneous time-varying random stepsizes satisfying standard conditions such as ∑α_k=∞ and ∑α_k²<∞ (or their random analogs). The manuscript must explicitly show that multiplication by the (possibly constant or time-varying) attenuation factor does not violate these summability properties or introduce unaccounted bias in the encrypted updates; without this, the interface between encryption mitigation and the stochastic approximation argument is not load-bearing.
minor comments (2)
  1. [Abstract / Introduction] The abstract and introduction would benefit from a brief proof sketch or explicit statement of the stepsize conditions and attenuation factor bounds to make the central claims more transparent.
  2. [Numerical simulations] Numerical simulations section: Include quantitative metrics on privacy leakage (e.g., mutual information or reconstruction error) alongside convergence plots to substantiate the privacy claims.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the detailed and constructive review. The single major comment identifies a point where the interface between the attenuation factor and the stochastic approximation argument requires explicit verification. We address it below and will incorporate the requested clarifications in the revised manuscript.

read point-by-point responses

  1. Referee: Convergence analysis (likely §4 or Theorem on a.s. convergence): The claim that the attenuation factor mitigates quantization error while preserving almost-sure convergence relies on the heterogeneous time-varying random stepsizes satisfying standard conditions such as ∑α_k=∞ and ∑α_k²<∞ (or their random analogs). The manuscript must explicitly show that multiplication by the (possibly constant or time-varying) attenuation factor does not violate these summability properties or introduce unaccounted bias in the encrypted updates; without this, the interface between encryption mitigation and the stochastic approximation argument is not load-bearing.

    Authors: We agree that an explicit verification is necessary to make the proof self-contained. In the revised version we will insert a short auxiliary result (new Lemma 4.3) showing the following: (i) if the original step-size sequence {α_k} satisfies ∑α_k=∞ and ∑α_k²<∞ almost surely, and the attenuation factor β_k ∈ (0,1] is either constant or satisfies β_k ≥ β_min >0 with probability 1, then the attenuated sequence {β_k α_k} inherits the same summability properties; (ii) the quantization error term, after multiplication by β_k, remains a martingale-difference sequence whose second-moment bound is still summable, introducing no additional asymptotic bias. The proof of the main almost-sure convergence theorem then proceeds verbatim with the attenuated updates. We will also add a brief remark clarifying that β_k is chosen independently of the random step-sizes and does not depend on the encrypted values, thereby preserving the required independence properties. revision: yes

Circularity Check

0 steps flagged

No significant circularity; convergence analysis relies on standard stochastic approximation conditions independent of the encryption mitigation

full rationale

The paper presents a distributed SGD algorithm augmented with Paillier encryption and an attenuation factor for quantization error. The abstract and reader's summary indicate that almost-sure convergence is asserted to follow from the heterogeneous time-varying random stepsizes satisfying the usual summability conditions (∑α_k=∞ and ∑α_k²<∞), with the attenuation factor chosen to preserve those properties. No equations are shown that define a key quantity in terms of itself, rename a fitted parameter as a prediction, or reduce the central claim to a self-citation chain. The attenuation factor is introduced as a design choice whose compatibility with convergence must be verified, but this is an assumption to be checked rather than a definitional loop. The derivation chain therefore remains self-contained against external benchmarks such as classical stochastic approximation theorems.

Axiom & Free-Parameter Ledger

2 free parameters · 2 axioms · 0 invented entities

Review performed on abstract only; full derivation, assumptions, and parameter choices are unavailable. The ledger therefore records only the minimal elements extractable from the abstract.

free parameters (2)
  • attenuation factor
    Introduced to offset quantization error from encryption; its functional form and selection rule are not specified in the abstract.
  • heterogeneous time-varying random stepsizes
    Each agent uses its own random sequence; the distributions or bounds required for convergence are not given.
axioms (2)
  • domain assumption The underlying optimization problem satisfies the conditions under which heterogeneous random stepsizes yield almost-sure convergence.
    Standard background assumption in stochastic optimization invoked to support the convergence claim.
  • domain assumption Paillier encryption plus attenuation preserves the unbiasedness or convergence properties of the stochastic gradients up to controllable error.
    Implicit in the claim that privacy is maintained while convergence still occurs.

pith-pipeline@v0.9.0 · 5421 in / 1638 out tokens · 49962 ms · 2026-05-09T21:06:41.523739+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

84 extracted references · 84 canonical work pages

  1. [1]

    2024 , eprint =

    Lastname, Firstname and Coauthor, Secondname , title =. 2024 , eprint =

    work page 2024

  2. [2]

    Secure and Privacy-Preserving Consensus , year=

    Ruan, Minghao and Gao, Huan and Wang, Yongqiang , journal=. Secure and Privacy-Preserving Consensus , year=

    work page

  3. [3]

    Byzantine-Resilient Decentralized Stochastic Optimization With Robust Aggregation Rules , year=

    Wu, Zhaoxian and Chen, Tianyi and Ling, Qing , journal=. Byzantine-Resilient Decentralized Stochastic Optimization With Robust Aggregation Rules , year=

    work page

  4. [4]

    Stochastic distributed learning with gradient quantization and double-variance reduction , publisher =

    Samuel Horv. Stochastic distributed learning with gradient quantization and double-variance reduction , publisher =. Optimization Methods and Software , number =. 2023 , volume =

    work page 2023

  5. [5]

    Decentralized Gossip-Based Stochastic Bilevel Optimization over Communication Networks , volume =

    Yang, Shuoguang and Zhang, Xuezhou and Wang, Mengdi , booktitle =. Decentralized Gossip-Based Stochastic Bilevel Optimization over Communication Networks , volume =

    work page

  6. [6]

    Quantization Enabled Privacy Protection in Decentralized Stochastic Optimization , year=

    Wang, Yongqiang and Başar, Tamer , journal=. Quantization Enabled Privacy Protection in Decentralized Stochastic Optimization , year=

    work page

  7. [7]

    Privacy-Preserving Distributed Online Stochastic Optimization With Time-Varying Distributions , year=

    Wang, Haojun and Liu, Kun and Han, Dongyu and Chai, Senchun and Xia, Yuanqing , journal=. Privacy-Preserving Distributed Online Stochastic Optimization With Time-Varying Distributions , year=

    work page

  8. [8]

    Differential Privacy in Distributed Optimization With Gradient Tracking , year=

    Huang, Lingying and Wu, Junfeng and Shi, Dawei and Dey, Subhrakanti and Shi, Ling , journal=. Differential Privacy in Distributed Optimization With Gradient Tracking , year=

    work page

  9. [9]

    Privacy-Preserving Average Consensus via State Decomposition , year=

    Wang, Yongqiang , journal=. Privacy-Preserving Average Consensus via State Decomposition , year=

    work page

  10. [10]

    Privacy-Preserving Consensus for Multi-Agent Systems via Node Decomposition Strategy , year=

    Wang, Yaqi and Lu, Jianquan and Zheng, Wei Xing and Shi, Kaibo , journal=. Privacy-Preserving Consensus for Multi-Agent Systems via Node Decomposition Strategy , year=

    work page

  11. [11]

    Local Differential Privacy for Decentralized Online Stochastic Optimization With Guaranteed Optimality and Convergence Speed , year=

    Chen, Ziqin and Wang, Yongqiang , journal=. Local Differential Privacy for Decentralized Online Stochastic Optimization With Guaranteed Optimality and Convergence Speed , year=

    work page

  12. [12]

    Convergence in High Probability of Distributed Stochastic Gradient Descent Algorithms , year=

    Lu, Kaihong and Wang, Hongxia and Zhang, Huanshui and Wang, Long , journal=. Convergence in High Probability of Distributed Stochastic Gradient Descent Algorithms , year=

    work page

  13. [13]

    Convergence Rates for Distributed Stochastic Optimization Over Random Networks , year=

    Jakovetic, Dusan and Bajovic, Dragana and Sahu, Anit Kumar and Kar, Soummya , booktitle=. Convergence Rates for Distributed Stochastic Optimization Over Random Networks , year=

    work page

  14. [14]

    , journal=

    Pu, Shi and Olshevsky, Alex and Paschalidis, Ioannis Ch. , journal=. A Sharp Estimate on the Transient Time of Distributed Stochastic Gradient Descent , year=

    work page

  15. [15]

    Proceedings of the International Conference on Neural Information Processing Systems , pages=

    Can decentralized algorithms outperform centralized algorithms? a case study for decentralized parallel stochastic gradient descent , author=. Proceedings of the International Conference on Neural Information Processing Systems , pages=

    work page

  16. [16]

    Enabling Privacy-Preservation in Decentralized Optimization , year=

    Zhang, Chunlei and Wang, Yongqiang , journal=. Enabling Privacy-Preservation in Decentralized Optimization , year=

    work page

  17. [17]

    A Distributed Stochastic Gradient Tracking Method , year=

    Pu, Shi and Nedić, Angelia , booktitle=. A Distributed Stochastic Gradient Tracking Method , year=

    work page

  18. [18]

    Tighter Analysis for Decentralized Stochastic Gradient Method: Impact of Data Homogeneity , year=

    Li, Qiang and Wai, Hoi-To , journal=. Tighter Analysis for Decentralized Stochastic Gradient Method: Impact of Data Homogeneity , year=

    work page

  19. [19]

    Distributed Stochastic Averaging Gradient Algorithm over Directed Graph , year=

    Zhao, Jiahong and Gao, Wenhua , booktitle=. Distributed Stochastic Averaging Gradient Algorithm over Directed Graph , year=

    work page

  20. [20]

    and Kar, Soummya , journal=

    Xin, Ran and Khan, Usman A. and Kar, Soummya , journal=. An Improved Convergence Analysis for Decentralized Online Stochastic Non-Convex Optimization , year=

    work page

  21. [21]

    On the Decentralized Stochastic Gradient Descent With Markov Chain Sampling , year=

    Sun, Tao and Li, Dongsheng and Wang, Bao , journal=. On the Decentralized Stochastic Gradient Descent With Markov Chain Sampling , year=

    work page

  22. [22]

    On the Convergence of Decentralized Stochastic Gradient Descent With Biased Gradients , year=

    Jiang, Yiming and Kang, Helei and Liu, Jinlan and Xu, Dongpo , journal=. On the Convergence of Decentralized Stochastic Gradient Descent With Biased Gradients , year=

    work page

  23. [23]

    2019 , volume=

    Zhang, Chunlei and Ahmad, Muaz and Wang, Yongqiang , journal=. 2019 , volume=

    work page 2019

  24. [24]

    Privacy-Preserving Collaborative Estimation for Networked Vehicles With Application to Collaborative Road Profile Estimation , year=

    Gao, Huan and Li, Zhaojian and Wang, Yongqiang , journal=. Privacy-Preserving Collaborative Estimation for Networked Vehicles With Application to Collaborative Road Profile Estimation , year=

    work page

  25. [25]

    , booktitle=

    Gade, Shripad and Vaidya, Nitin H. , booktitle=. Privacy-Preserving Distributed Learning via Obfuscated Stochastic Gradients , year=

    work page

  26. [26]

    and Qi, Yuan , journal=

    Yan, Feng and Sundaram, Shreyas and Vishwanathan, S.V.N. and Qi, Yuan , journal=. Distributed Autonomous Online Learning: Regrets and Intrinsic Privacy-Preserving Properties , year=

    work page

  27. [27]

    How to generate and exchange secrets , year=

    Yao, Andrew Chi-Chih , booktitle=. How to generate and exchange secrets , year=

    work page

  28. [28]

    Paillier, Pascal , title=. Proc. Int. Conf. Theory Appl. Cryptogr. Techn. , year=

    work page

  29. [29]

    Brendan and Mironov, Ilya and Talwar, Kunal and Zhang, Li , title =

    Abadi, Martin and Chu, Andy and Goodfellow, Ian and McMahan, H. Brendan and Mironov, Ilya and Talwar, Kunal and Zhang, Li , title =. 2016 , booktitle =

    work page 2016

  30. [30]

    Sarwate , title =

    Kamalika Chaudhuri and Claire Monteleoni and Anand D. Sarwate , title =. Journal of Machine Learning Research , year =

    work page

  31. [31]

    Calibrating Noise to Sensitivity in Private Data Analysis

    Dwork, Cynthia and McSherry, Frank and Nissim, Kobbi and Smith, Adam. Calibrating Noise to Sensitivity in Private Data Analysis. Theory of Cryptography. 2006

    work page 2006

  32. [32]

    Deep Leakage from Gradients , pages =

    Zhu, Ligeng and Liu, Zhijian and Han, Song , booktitle =. Deep Leakage from Gradients , pages =

    work page

  33. [33]

    Exploiting Unintended Feature Leakage in Collaborative Learning , year=

    Melis, Luca and Song, Congzheng and De Cristofaro, Emiliano and Shmatikov, Vitaly , booktitle=. Exploiting Unintended Feature Leakage in Collaborative Learning , year=

    work page

  34. [34]

    Privacy-preserving deep learning , year=

    Shokri, Reza and Shmatikov, Vitaly , booktitle=. Privacy-preserving deep learning , year=

    work page

  35. [35]

    , journal=

    Xi, Chenguang and Xin, Ran and Khan, Usman A. , journal=. ADD-OPT: Accelerated Distributed Directed Optimization , year=

    work page

  36. [36]

    Augmented distributed gradient methods for multi-agent optimization under uncoordinated constant stepsizes , year=

    Xu, Jinming and Zhu, Shanying and Soh, Yeng Chai and Xie, Lihua , booktitle=. Augmented distributed gradient methods for multi-agent optimization under uncoordinated constant stepsizes , year=

    work page

  37. [37]

    Achieving Geometric Convergence for Distributed Optimization Over Time-Varying Graphs , journal =

    Nedi\'. Achieving Geometric Convergence for Distributed Optimization Over Time-Varying Graphs , journal =

    work page

  38. [38]

    Yuan, Kun and Ling, Qing and Yin, Wotao , title =. SIAM J. Optim. , volume =

    work page

  39. [39]

    Distributed Stochastic Gradient Method for Non-Convex Problems with Applications in Supervised Learning , year=

    George, Jemin and Yang, Tao and Bai, He and Gurram, Prudhvi , booktitle=. Distributed Stochastic Gradient Method for Non-Convex Problems with Applications in Supervised Learning , year=

    work page

  40. [40]

    Convergence of a Multi-Agent Projected Stochastic Gradient Algorithm for Non-Convex Optimization , year=

    Bianchi, Pascal and Jakubowicz, Jérémie , journal=. Convergence of a Multi-Agent Projected Stochastic Gradient Algorithm for Non-Convex Optimization , year=

    work page

  41. [41]

    2018 IEEE Conference on Decision and Control (CDC) , title=

    Pu, Shi and Nedi\'. 2018 IEEE Conference on Decision and Control (CDC) , title=. 2018 , volume=

    work page 2018

  42. [42]

    IEEE Transactions on Automatic Control , volume=

    Distributed subgradient methods for multi-agent optimization , author=. IEEE Transactions on Automatic Control , volume=. 2009 , publisher=

    work page 2009

  43. [43]

    and Bertsekas, D

    Tsitsiklis, J. and Bertsekas, D. and Athans, M. , journal=. Distributed asynchronous deterministic and stochastic gradient optimization algorithms , year=

    work page

  44. [44]

    Sundhar and Nedic, A

    Ram, S. Sundhar and Nedic, A. and Veeravalli, V. V. , booktitle=. Distributed subgradient projection algorithm for convex optimization , year=

    work page

  45. [45]

    Distributed Control of Robotic Networks: A Mathematical Approach to Motion Coordination Algorithms , urldate =

    Francesco Bullo and Jorge Cortés and Sonia Martínez , publisher =. Distributed Control of Robotic Networks: A Mathematical Approach to Motion Coordination Algorithms , urldate =

    work page

  46. [46]

    A Stochastic Approximation Method , urldate =

    Herbert Robbins and Sutton Monro , journal =. A Stochastic Approximation Method , urldate =

    work page

  47. [47]

    and Lawlor, Sean and Rabbat, Michael G

    Tsianos, Konstantinos I. and Lawlor, Sean and Rabbat, Michael G. , booktitle=. Consensus-based distributed optimization: Practical issues and applications in large-scale machine learning , year=

    work page

  48. [48]

    IEEE Trans

    Wang, Yongqiang and Nedi\'. IEEE Trans. Autom. Control , title=. 2024 , volume=

    work page 2024

  49. [49]

    IEEE Trans

    Nedi\'. IEEE Trans. Autom. Control , title=. 2009 , volume=

    work page 2009

  50. [50]

    Vincent , journal=

    Wang, Yongqiang and Poor, H. Vincent , journal=. Decentralized Stochastic Optimization With Inherent Privacy Protection , year=

    work page

  51. [51]

    IEEE Transactions on Automatic Control , volume =

    Decentralized Gradient Methods With Time-Varying Uncoordinated Stepsizes: Convergence Analysis and Privacy Design , author =. IEEE Transactions on Automatic Control , volume =

    work page

  52. [52]

    IEEE Trans

    Nedi\'. IEEE Trans. Autom. Control , title=. 2016 , volume=

    work page 2016

  53. [53]

    Distributed optimization in sensor networks , author=. Proc. ACM Inf. Process. Sensor Networks. , pages=

    work page

  54. [54]

    International Journal of Robust and Nonlinear Control , year=

    Privacy-Preserving Optimization Algorithm for Distributed Energy Management Over Time-Varying Graphs: A State Decomposition Method , author=. International Journal of Robust and Nonlinear Control , year=

    work page

  55. [55]

    IEEE Trans

    Hybrid random/deterministic parallel algorithms for convex and nonconvex big data optimization , author=. IEEE Trans. Signal Process. , volume=. 2015 , publisher=

    work page 2015

  56. [56]

    Automatica , volume =

    Dynamics based privacy preservation in decentralized optimization , author =. Automatica , volume =. 2023 , publisher =

    work page 2023

  57. [57]

    Oblivious Multi-Party machine learning on trusted processors , author=. Proc. 25th USENIX Conf. Secur. Symp. , pages=

    work page

  58. [58]

    Slalom: Fast, verifiable and private execution of neural networks in trusted hardware , author=. Proc. ICLR , pages=

    work page

  59. [59]

    Abadi, M., Chu, A., Goodfellow, I., McMahan, H.B., Mironov, I., Talwar, K., and Zhang, L. (2016). Deep learning with differential privacy. In Proc. ACM SIGSAC Conf. Comput. Commun. Security, 308–318

    work page 2016

  60. [60]

    and Jakubowicz, J

    Bianchi, P. and Jakubowicz, J. (2013). Convergence of a multi-agent projected stochastic gradient algorithm for non-convex optimization. IEEE Trans. Autom. Control, 58(2), 391--405

    work page 2013

  61. [61]

    Daneshmand, A., Facchinei, F., Kungurtsev, V., and Scutari, G. (2015). Hybrid random/deterministic parallel algorithms for convex and nonconvex big data optimization. IEEE Trans. Signal Process., 63(15), 3914--3929

    work page 2015

  62. [62]

    and Vaidya, N.H

    Gade, S. and Vaidya, N.H. (2018). Privacy-preserving distributed learning via obfuscated stochastic gradients. In IEEE Conf. Decis. Control, 184--191

    work page 2018

  63. [63]

    Gao, H., Wang, Y., and Nedi \'c , A. (2023). Dynamics based privacy preservation in decentralized optimization. Automatica, 151, 110878

    work page 2023

  64. [64]

    Huang, L., Wu, J., Shi, D., Dey, S., and Shi, L. (2024). Differential privacy in distributed optimization with gradient tracking. IEEE Trans. Autom. Control, 69(9), 5727--5742

    work page 2024

  65. [65]

    Jiang, Y., Kang, H., Liu, J., and Xu, D. (2025). On the convergence of decentralized stochastic gradient descent with biased gradients. IEEE Trans. Signal Process., 73, 549--558

    work page 2025

  66. [66]

    and Wai, H.T

    Li, Q. and Wai, H.T. (2025). Tighter analysis for decentralized stochastic gradient method: Impact of data homogeneity. IEEE Trans. Autom. Control, 70(7), 4703--4718

    work page 2025

  67. [67]

    and Olshevsky, A

    Nedi\' c , A. and Olshevsky, A. (2016). Stochastic gradient-push for strongly convex functions on time-varying directed graphs. IEEE Trans. Autom. Control, 61(12), 3936--3947

    work page 2016

  68. [68]

    and Ozdaglar, A

    Nedi\' c , A. and Ozdaglar, A. (2009). Distributed subgradient methods for multi-agent optimization. IEEE Trans. Autom. Control, 54(1), 48--61

    work page 2009

  69. [69]

    Ohrimenko, O., Schuster, F., Fournet, C., Mehta, A., Nowozin, S., Vaswani, K., and Costa, M. (2016). Oblivious multi-party machine learning on trusted processors. In Proc. 25th USENIX Conf. Secur. Symp., 619--636

    work page 2016

  70. [70]

    Paillier, P. (1999). Public-key cryptosystems based on composite degree residuosity classes. In Proc. Int. Conf. Theory Appl. Cryptogr. Techn., 223--238

    work page 1999

  71. [71]

    and Nedić, A

    Pu, S. and Nedić, A. (2018). A distributed stochastic gradient tracking method. In IEEE Conf. Decis. Control, 963--968

    work page 2018

  72. [72]

    and Nowak, R

    Rabbat, M. and Nowak, R. (2004). Distributed optimization in sensor networks. In Proc. ACM Inf. Process. Sensor Networks., 20--27

    work page 2004

  73. [73]

    and Monro, S

    Robbins, H. and Monro, S. (1951). A stochastic approximation method. Ann. Math. Statist., 22(3), 400--407

    work page 1951

  74. [74]

    Ruan, M., Gao, H., and Wang, Y. (2019). Secure and privacy-preserving consensus. IEEE Trans. Autom. Control, 64(10), 4035--4049

    work page 2019

  75. [75]

    and Boneh, D

    Tramer, F. and Boneh, D. (2018). Slalom: Fast, verifiable and private execution of neural networks in trusted hardware. In Proc. ICLR, 1--19

    work page 2018

  76. [76]

    Tsianos, K.I., Lawlor, S., and Rabbat, M.G. (2012). Consensus-based distributed optimization: Practical issues and applications in large-scale machine learning. In Proc. 50th Annu. Allerton Conf. Commun., Control, Comput., 1543--1550

    work page 2012

  77. [77]

    Tsitsiklis, J., Bertsekas, D., and Athans, M. (1986). Distributed asynchronous deterministic and stochastic gradient optimization algorithms. IEEE Trans. Autom. Control, 31(9), 803--812

    work page 1986

  78. [78]

    and Başar, T

    Wang, Y. and Başar, T. (2023). Quantization enabled privacy protection in decentralized stochastic optimization. IEEE Trans. Autom. Control, 68(7), 4038--4052

    work page 2023

  79. [79]

    and Nedi\' c , A

    Wang, Y. and Nedi\' c , A. (2024). Tailoring gradient methods for differentially private distributed optimization. IEEE Trans. Autom. Control, 69(2), 872--887

    work page 2024

  80. [80]

    and Poor, H.V

    Wang, Y. and Poor, H.V. (2023). Decentralized stochastic optimization with inherent privacy protection. IEEE Trans. Autom. Control, 68(4), 2293--2308

    work page 2023

Showing first 80 references.