pith. machine review for the scientific record. sign in

arxiv: 2605.07841 · v1 · submitted 2026-05-08 · 💻 cs.LG · cs.AI· cs.DC

Recognition: 2 theorem links

· Lean Theorem

mathsf{VISTA}: Decentralized Machine Learning in Adversary Dominated Environments

Hanzaleh Akbari Nodehi, Mohammad Ali Maddah-Ali, Parsa Moradi, Soheil Mohajer

Authors on Pith no claims yet

Pith reviewed 2026-05-11 02:25 UTC · model grok-4.3

classification 💻 cs.LG cs.AIcs.DC
keywords decentralized machine learningadversarial robustnessincentive mechanismsadaptive algorithmsconvergence analysisSGD
0
0 comments X

The pith

Adaptive threshold tuning lets decentralized SGD converge asymptotically despite adversary majority.

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

Decentralized machine learning often outsources gradient computations to untrusted nodes, but most robust methods assume an honest majority of workers. This paper introduces an incentive-oriented framework where reports are only accepted if they are mutually consistent up to a threshold, making adversaries rational agents who weigh the benefit of corrupting estimates against the risk of rejection and lost rewards. The proposed VISTA algorithm adaptively adjusts this threshold based on optimization history to allow faster early progress while eventually enforcing stricter consistency. Numerical results indicate better convergence than fixed thresholds, and theoretical analysis shows that asymptotic convergence matches that of standard SGD. This matters for practical systems where controlling a majority of nodes is common and honest-majority assumptions do not hold.

Core claim

We propose VISTA, an adaptive algorithm that tunes the acceptance threshold using the optimization history. With suitable incentive-aware adaptation, adversary-dominated decentralized learning can retain the asymptotic convergence behavior of standard SGD without relying on an honest majority.

What carries the argument

The incentive-oriented acceptance rule combined with the VISTA adaptive threshold tuner, which uses past optimization progress to decide how permissive to be with report consistency.

If this is right

  • Adversary-dominated settings achieve the same asymptotic convergence rate as standard SGD.
  • Dynamic threshold adaptation outperforms static rules by balancing speed and accuracy over iterations.
  • Adversaries are forced to act rationally, limiting the damage they can cause without detection.
  • Long-horizon iterative optimization benefits from history-dependent rules rather than one-shot decisions.

Where Pith is reading between the lines

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

  • Extensions could include applying similar adaptation to non-convex optimization or federated learning with partial participation.
  • Empirical validation in real distributed systems with actual reward mechanisms would strengthen the incentive claims.
  • The approach opens questions about optimal threshold schedules for different adversary strategies.

Load-bearing premise

Adversaries act as rational agents balancing the increase in estimation error against the risk of report rejection and loss of reward, without the adaptive rule introducing instabilities.

What would settle it

If simulations with a majority of adversarial workers show that VISTA fails to converge or converges significantly slower than standard SGD, the claim would be falsified.

Figures

Figures reproduced from arXiv: 2605.07841 by Hanzaleh Akbari Nodehi, Mohammad Ali Maddah-Ali, Parsa Moradi, Soheil Mohajer.

Figure 1
Figure 1. Figure 1: Fig. (a) shows a DC outsourcing the computation of L(W) to worker nodes due to limited resources. Honest workers follow the protocol but may return bounded noisy outputs, while adversaries may strategically inject arbitrary noise. We focus on the adversary dominated regime. Fig. (b) shows a two-node game-of-coding instance [28–32], with one honest and one adversarial node. The honest node returns Yh = L(W)… view at source ↗
Figure 2
Figure 2. Figure 2: Iterative game of coding framework. Since the DC does not have sufficient resources to evaluate ∇L(Wt−1) on its own, at each round t it broadcasts the current model parameter Wt−1 and the acceptance parameter ηt to the worker nodes. Each node i returns a noisy gradient report Yi,t = ∇L(Wt−1) + Ni,t. For honest workers, we have ∥Ni,t∥2 ≤ ∆, while adversarial workers choose their noise strategically accordin… view at source ↗
Figure 3
Figure 3. Figure 3: A larger constant η enables faster initial progress through more frequent updates, but leads to higher error by admitting more adversarial noise. Smaller ones are more reliable but progress slowly due to frequent rejections. VISTA adapts ηt over time, using larger ηt early and smaller later to achieve best performance. This paper resolves this dilemma by proposing an algorithm that adaptively selects the a… view at source ↗
Figure 4
Figure 4. Figure 4: Training loss on MNIST for LeNet. 4 Experimental Results In this section, we empirically evaluate VISTA. We use prior works on game of coding [28–32] to characterize pηt , σ 2 ηt , and the estimator est(·). Unless stated otherwise, we consider n = 2 nodes with |K| = |Q| = 1, and use adversarial utilities of the form UAD = log(MSE) + λ log(PA). All implementation details are provided in Appendix B. We also … view at source ↗
Figure 5
Figure 5. Figure 5: Equilibrium characterization for the three-dimensional experiment in Section 4. The [PITH_FULL_IMAGE:figures/full_fig_p020_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Convergence in the one-dimensional case. Larger values of η improve early progress but worsen final accuracy, while the proposed scheme achieves the best performance. one honest node and one adversarial node, has been characterized in [29]. This is the vector-valued setting used in our experiments in Section 4. The general vector-valued case with more than two nodes remains open; once the corresponding equ… view at source ↗
Figure 7
Figure 7. Figure 7: Dynamic evolution of the acceptance threshold ηt across the experiments. (a) One-dimensional experiment. (b) Three-dimensional experiment. In both cases, the proposed scheme adapts ηt over time, starting from a larger value and gradually tightening the acceptance rule. C Sensitivity to the Hyperparameter c In this section, we study the role of the hyperparameter c in VISTA. Recall that in Line 7 of Algorit… view at source ↗
Figure 8
Figure 8. Figure 8: Dynamic evolution of the acceptance threshold ηt across the experiments. (a) LeNet on MNIST. (b) ResNet-18 on CIFAR-10. In both cases, the proposed scheme adapts ηt over time, starting from a larger value and gradually tightening the acceptance rule. as a signal-to-noise tolerance parameter: smaller values are more conservative, while larger values are more aggressive. Figures 9 and 10 demonstrates the rob… view at source ↗
Figure 9
Figure 9. Figure 9: Sensitivity of VISTA to the hyperparameter c. (a) Mean squared gradient in the one-dimensional experiment. Several values of c achieve strong convergence, showing that the method is not sensitive to a single value. (b) Training loss for ResNet-18 on CIFAR-10. Larger values of c improve performance in this high-dimensional setting by keeping the algorithm in the adaptive regime for longer. (a) One-dimension… view at source ↗
Figure 10
Figure 10. Figure 10: Evolution of the acceptance threshold ηt for different values of c. (a) In the one-dimensional experiment, smaller c values reduce ηt more aggressively, while larger values keep the algorithm adaptive for longer. (b) In the ResNet-18 experiment, increasing c leads to larger thresholds during the early phase, which enables more frequent accepted updates before the threshold is gradually tightened. 24 [PIT… view at source ↗
read the original abstract

Decentralized machine learning often relies on outsourcing computations, such as gradient evaluations, to untrusted worker nodes. Existing robust aggregation methods can mitigate malicious behavior under honest-majority assumptions, but may fail when adversaries control a majority of the workers. We study this adversary-dominated setting through an incentive-oriented framework in which reports are accepted and rewarded only when they are mutually consistent up to a threshold. This turns the adversary from a pure saboteur into a rational agent that trades off increasing estimation error against the risk of rejection and loss of reward. We consider iterative optimization under this model. Unlike one-shot computation, iterative learning requires long-horizon decisions: permissive acceptance rules enable faster early progress but admit more adversarial corruption, while strict rules improve estimation accuracy but cause frequent rejections. We propose \mathsf{VISTA}, an adaptive algorithm that tunes the acceptance threshold using the optimization history. Numerical results show that \mathsf{VISTA} improves convergence over static thresholds. We also provide a rigorous convergence analysis showing that, with suitable incentive-aware adaptation, adversary-dominated decentralized learning can retain the asymptotic convergence behavior of standard SGD without relying on an honest majority.

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

2 major / 3 minor

Summary. The paper proposes VISTA, an adaptive algorithm for decentralized machine learning when adversaries control a majority of workers. It introduces an incentive framework in which reports are accepted and rewarded only if mutually consistent up to a tunable threshold, modeling adversaries as rational agents balancing estimation error against rejection risk. VISTA adjusts the threshold dynamically using optimization history to trade off early progress against corruption. The authors claim numerical improvements over static thresholds and provide a rigorous convergence analysis asserting that, under suitable adaptation, the method retains the asymptotic convergence rate of standard SGD without requiring an honest majority.

Significance. If the central convergence claim holds, the work is significant for relaxing the honest-majority assumption prevalent in robust aggregation literature for decentralized and federated learning. The incentive-oriented modeling and history-based adaptation offer a fresh perspective on long-horizon decisions in iterative optimization under adversarial incentives. Credit is due for supplying both a theoretical analysis and numerical validation; the absence of free parameters in the core model and the explicit focus on asymptotic behavior are strengths.

major comments (2)
  1. [§5] §5 (Convergence Analysis), main theorem: The proof that asymptotic SGD-like convergence is retained assumes the adaptive threshold keeps the effective bias from accepted gradients controlled. However, when adversaries exceed 50% and coordinate on identical but biased reports, they form a large consistent cluster; the history-based relaxation rule can incrementally raise the threshold, admitting persistent bias without an explicit bound on the resulting deviation from the true gradient. This is load-bearing for the claim that no honest majority is required.
  2. [§3] §3 (Model), rational-adversary definition: The assumption that adversaries rationally trade estimation error against rejection risk does not address coordinated equilibria. When a majority colludes on a common biased value while remaining internally consistent, the mutual-consistency check accepts the cluster; the paper provides no equilibrium analysis or bound showing that the adaptive rule still forces the bias to decay at the SGD rate.
minor comments (3)
  1. [Abstract] Abstract: The phrase 'rigorous convergence analysis' appears without a forward reference to the specific theorem or the precise assumptions (e.g., bounded variance, Lipschitz constants) under which the SGD rate is recovered.
  2. [Numerical results] Numerical results section: Convergence plots are presented without reporting the number of independent runs, standard deviations, or confidence intervals, making it difficult to judge whether the observed improvement over static thresholds is statistically reliable.
  3. [Algorithm description] Notation: The symbol for the adaptive threshold is introduced without an explicit recurrence or update rule in the main text; readers must infer the exact dependence on optimization history from the algorithm box alone.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the detailed and constructive report. The two major comments identify important points regarding the handling of coordinated adversaries in the convergence analysis and the modeling assumptions. We address each below and outline targeted revisions that strengthen the claims without altering the core contributions.

read point-by-point responses

  1. Referee: [§5] §5 (Convergence Analysis), main theorem: The proof that asymptotic SGD-like convergence is retained assumes the adaptive threshold keeps the effective bias from accepted gradients controlled. However, when adversaries exceed 50% and coordinate on identical but biased reports, they form a large consistent cluster; the history-based relaxation rule can incrementally raise the threshold, admitting persistent bias without an explicit bound on the resulting deviation from the true gradient. This is load-bearing for the claim that no honest majority is required.

    Authors: We appreciate the referee drawing attention to this aspect of the proof. The main theorem in §5 establishes asymptotic SGD-like rates under the adaptive rule by showing that the threshold relaxes only when historical consistency aligns with objective progress; persistent bias from a coordinated cluster would stall descent and thereby cap further relaxation. Nevertheless, we agree that an explicit bias bound for the >50% coordinated case would make the argument more transparent. In the revision we will insert a supporting lemma in §5 that derives a uniform bound on the deviation of accepted gradients under majority-consistent attacks, confirming that the adaptation still forces the bias term to vanish at the required rate. This addition directly addresses the load-bearing concern while preserving the existing proof structure. revision: partial

  2. Referee: [§3] §3 (Model), rational-adversary definition: The assumption that adversaries rationally trade estimation error against rejection risk does not address coordinated equilibria. When a majority colludes on a common biased value while remaining internally consistent, the mutual-consistency check accepts the cluster; the paper provides no equilibrium analysis or bound showing that the adaptive rule still forces the bias to decay at the SGD rate.

    Authors: The referee correctly observes that §3 models individual rational trade-offs and does not contain a dedicated equilibrium analysis for colluding majorities. While the incentive structure and history-based adaptation are intended to deter sustained large biases (because non-progress halts threshold relaxation), we acknowledge the absence of an explicit game-theoretic treatment of coordinated equilibria. In the revised manuscript we will add a short subsection to §3 that sketches the relevant equilibrium considerations and shows, via the same progress-dependent relaxation mechanism used in the convergence proof, that rational colluders cannot maintain a fixed positive bias indefinitely without violating the long-horizon reward objective. This discussion will be cross-referenced to the new lemma in §5. revision: partial

Circularity Check

0 steps flagged

No significant circularity; convergence analysis builds on external SGD assumptions

full rationale

The paper's derivation chain centers on an adaptive threshold rule motivated by long-horizon optimization trade-offs, followed by a claimed rigorous convergence analysis that retains standard SGD asymptotics. No equations or steps in the provided text reduce the target convergence rate to a fitted parameter, self-defined quantity, or self-citation chain. The analysis is presented as extending standard SGD assumptions under the incentive model rather than tautologically re-deriving them from the adaptation rule itself. Numerical results are experimental and do not substitute for the analytic claim. This is the common case of a self-contained derivation against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

The central claim rests on modeling adversaries as rational reward-maximizers and on the existence of an adaptive rule that preserves SGD asymptotics; these are introduced without independent external validation in the provided abstract.

axioms (1)
  • domain assumption Standard SGD convergence assumptions (bounded variance, unbiased gradients, appropriate step sizes)
    The analysis claims to retain asymptotic SGD behavior, which implicitly relies on these classical conditions.
invented entities (1)
  • Rational adversary that trades estimation error against rejection risk no independent evidence
    purpose: To convert pure sabotage into a strategic decision that the acceptance threshold can exploit
    This modeling choice is introduced to enable the incentive framework; no external evidence is provided in the abstract.

pith-pipeline@v0.9.0 · 5519 in / 1142 out tokens · 30289 ms · 2026-05-11T02:25:31.521928+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

Reference graph

Works this paper leans on

94 extracted references · 94 canonical work pages · 3 internal anchors

  1. [1]

    Blockchain for deep learning: review and open challenges.Cluster Computing, 26(1):197–221, 2023

    Muhammad Shafay, Raja Wasim Ahmad, Khaled Salah, Ibrar Yaqoob, Raja Jayaraman, and Mohammed Omar. Blockchain for deep learning: review and open challenges.Cluster Computing, 26(1):197–221, 2023

    work page 2023

  2. [2]

    Survey on the convergence of machine learning and blockchain

    Shengwen Ding and Chenhui Hu. Survey on the convergence of machine learning and blockchain. InProceedings of SAI Intelligent Systems Conference, pages 170–189. Springer, 2022

    work page 2022

  3. [3]

    Blockchain meets machine learning: a survey

    Safak Kayikci and Taghi M Khoshgoftaar. Blockchain meets machine learning: a survey. Journal of Big Data, 11(1):1–29, 2024. 10

    work page 2024

  4. [4]

    Blockchain and machine learning: A critical review on security.Information, 14(5):295, 2023

    Hamed Taherdoost. Blockchain and machine learning: A critical review on security.Information, 14(5):295, 2023

    work page 2023

  5. [5]

    Blockchain technology and artificial intelligence together: a critical review on applications.Applied Sciences, 12(24):12948, 2022

    Hamed Taherdoost. Blockchain technology and artificial intelligence together: a critical review on applications.Applied Sciences, 12(24):12948, 2022

    work page 2022

  6. [6]

    Blockchain for AI: A disruptive integration

    Ruijiao Tian, Lanju Kong, Xinping Min, and Yunhao Qu. Blockchain for AI: A disruptive integration. In2022 IEEE 25th International Conference on Computer Supported Cooperative Work in Design (CSCWD), pages 938–943. IEEE, 2022

    work page 2022

  7. [7]

    Blockchain for AI: Review and open research challenges.IEEE Access, 7:10127–10149, 2019

    Khaled Salah, M Habib Ur Rehman, Nishara Nizamuddin, and Ala Al-Fuqaha. Blockchain for AI: Review and open research challenges.IEEE Access, 7:10127–10149, 2019

    work page 2019

  8. [8]

    VeriML: Enabling integrity assurances and fair payments for machine learning as a service.IEEE Transactions on Parallel and Distributed Systems, 32(10):2524–2540, 2021

    Lingchen Zhao, Qian Wang, Cong Wang, Qi Li, Chao Shen, and Bo Feng. VeriML: Enabling integrity assurances and fair payments for machine learning as a service.IEEE Transactions on Parallel and Distributed Systems, 32(10):2524–2540, 2021

    work page 2021

  9. [9]

    Blockchains cannot rely on honesty

    Jakub Sliwinski and Roger Wattenhofer. Blockchains cannot rely on honesty. InThe 19th International Conference on Autonomous Agents and Multiagent Systems (AAMAS 2020), 2019

    work page 2020

  10. [10]

    Fact and fiction: Challenging the honest majority assumption of permissionless blockchains

    Runchao Han, Zhimei Sui, Jiangshan Yu, Joseph Liu, and Shiping Chen. Fact and fiction: Challenging the honest majority assumption of permissionless blockchains. InProceedings of the 2021 ACM Asia Conference on Computer and Communications Security, pages 817–831, 2021

    work page 2021

  11. [11]

    Zero Cost

    Joshua S Gans and Hanna Halaburda. "Zero Cost" majority attacks on permissionless blockchains. Technical report, National Bureau of Economic Research, 2023

    work page 2023

  12. [12]

    Byzantine-resilient non- convex stochastic gradient descent.arXiv preprint arXiv:2012.14368, 2020

    Zeyuan Allen-Zhu, Faeze Ebrahimian, Jerry Li, and Dan Alistarh. Byzantine-resilient non- convex stochastic gradient descent.arXiv preprint arXiv:2012.14368, 2020

    work page arXiv 2012

  13. [13]

    Byzantine machine learning made easy by resilient averaging of momentums

    Sadegh Farhadkhani, Rachid Guerraoui, Nirupam Gupta, Rafael Pinot, and John Stephan. Byzantine machine learning made easy by resilient averaging of momentums. InInternational Conference on Machine Learning, pages 6246–6283. PMLR, 2022

    work page 2022

  14. [14]

    Optimal complexity in byzantine- robust distributed stochastic optimization with data heterogeneity.Journal of Machine Learning Research, 26(268):1–58, 2025

    Qiankun Shi, Jie Peng, Kun Yuan, Xiao Wang, and Qing Ling. Optimal complexity in byzantine- robust distributed stochastic optimization with data heterogeneity.Journal of Machine Learning Research, 26(268):1–58, 2025

    work page 2025

  15. [15]

    Learning from history for byzantine robust optimization

    Sai Praneeth Karimireddy, Lie He, and Martin Jaggi. Learning from history for byzantine robust optimization. InInternational Conference on Machine Learning, pages 5311–5319. PMLR, 2021

    work page 2021

  16. [16]

    Byzantine machine learning: A primer

    Rachid Guerraoui, Nirupam Gupta, and Rafael Pinot. Byzantine machine learning: A primer. ACM Computing Surveys, 56(7):1–39, 2024

    work page 2024

  17. [17]

    Byzantine-robust federated learning with optimal statistical rates

    Banghua Zhu, Lun Wang, Qi Pang, Shuai Wang, Jiantao Jiao, Dawn Song, and Michael I Jordan. Byzantine-robust federated learning with optimal statistical rates. InInternational Conference on Artificial Intelligence and Statistics, pages 3151–3178. PMLR, 2023

    work page 2023

  18. [18]

    Byzantine- robust learning on heterogeneous data via gradient splitting

    Yuchen Liu, Chen Chen, Lingjuan Lyu, Fangzhao Wu, Sai Wu, and Gang Chen. Byzantine- robust learning on heterogeneous data via gradient splitting. InInternational Conference on Machine Learning, pages 21404–21425. PMLR, 2023

    work page 2023

  19. [19]

    Byzantine-resilient stochastic gradient descent for distributed learning: A lipschitz-inspired coordinate-wise median approach

    Haibo Yang, Xin Zhang, Minghong Fang, and Jia Liu. Byzantine-resilient stochastic gradient descent for distributed learning: A lipschitz-inspired coordinate-wise median approach. In2019 IEEE 58th Conference on Decision and Control (CDC), pages 5832–5837. IEEE, 2019

    work page 2019

  20. [20]

    Detox: A redundancy-based framework for faster and more robust gradient aggregation.Advances in Neural Information Processing Systems, 32, 2019

    Shashank Rajput, Hongyi Wang, Zachary Charles, and Dimitris Papailiopoulos. Detox: A redundancy-based framework for faster and more robust gradient aggregation.Advances in Neural Information Processing Systems, 32, 2019

    work page 2019

  21. [21]

    Byzantine-robust federated learning through spatial-temporal analysis of local model updates

    Zhuohang Li, Luyang Liu, Jiaxin Zhang, and Jian Liu. Byzantine-robust federated learning through spatial-temporal analysis of local model updates. In2021 IEEE 27th International Conference on Parallel and Distributed Systems (ICPADS), pages 372–379. IEEE, 2021. 11

    work page 2021

  22. [22]

    Ro- bust distributed learning: Tight error bounds and breakdown point under data heterogeneity

    Youssef Allouah, Rachid Guerraoui, Nirupam Gupta, Rafael Pinot, and Geovani Rizk. Ro- bust distributed learning: Tight error bounds and breakdown point under data heterogeneity. Advances in Neural Information Processing Systems, 36:45744–45776, 2023

    work page 2023

  23. [23]

    Byzantine stochastic gradient descent.Advances in Neural Information Processing Systems, 31, 2018

    Dan Alistarh, Zeyuan Allen-Zhu, and Jerry Li. Byzantine stochastic gradient descent.Advances in Neural Information Processing Systems, 31, 2018

    work page 2018

  24. [24]

    El Mahdi El-Mhamdi, Sadegh Farhadkhani, Rachid Guerraoui, Arsany Guirguis, Lê-Nguyên Hoang, and Sébastien Rouault. Collaborative learning in the jungle (decentralized, byzantine, heterogeneous, asynchronous and nonconvex learning).Advances in Neural Information Processing Systems, 34:25044–25057, 2021

    work page 2021

  25. [25]

    The hidden vulnerability of distributed learning in Byzantium

    Rachid Guerraoui and Sébastien Rouault. The hidden vulnerability of distributed learning in Byzantium. InInternational Conference on Machine Learning, pages 3521–3530. PMLR, 2018

    work page 2018

  26. [26]

    Byzantine-robust distributed learning: Towards optimal statistical rates

    Dong Yin, Yudong Chen, Ramchandran Kannan, and Peter Bartlett. Byzantine-robust distributed learning: Towards optimal statistical rates. InInternational Conference on Machine Learning, pages 5650–5659. PMLR, 2018

    work page 2018

  27. [27]

    Machine learn- ing with adversaries: Byzantine tolerant gradient descent.Advances in Neural Information Processing Systems, 30, 2017

    Peva Blanchard, El Mahdi El Mhamdi, Rachid Guerraoui, and Julien Stainer. Machine learn- ing with adversaries: Byzantine tolerant gradient descent.Advances in Neural Information Processing Systems, 30, 2017

    work page 2017

  28. [28]

    Cadambe, and Mohammad Ali Maddah-Ali

    Hanzaleh Akbari Nodehi, Viveck R. Cadambe, and Mohammad Ali Maddah-Ali. Game of coding: Beyond honest-majority assumptions.IEEE Transactions on Information Theory (submitted), 2024. URLhttps://arxiv.org/abs/2401.16643

    work page arXiv 2024

  29. [29]

    Game of Coding for Vector-Valued Computations

    Hanzaleh Akbari Nodehi, Parsa Moradi, Soheil Mohajer, and Mohammad Ali Maddah-Ali. Game of coding for vector-valued computations.arXiv preprint arXiv:2602.04810, 2026

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  30. [30]

    Game of coding: Sybil resistant decentralized machine learning with minimal trust assumption,

    Hanzaleh Akbari Nodehi, Viveck R. Cadambe, and Mohammad Ali Maddah-Ali. Game of coding: Sybil resistant decentralized machine learning with minimal trust assumption.arXiv preprint, 2024. https://arxiv.org/abs/2410.05540

    work page arXiv 2024

  31. [31]

    Game of coding with an unknown adversary

    Hanzaleh Akbari Nodehi, Pejman Moradi, and Mohammad Ali Maddah-Ali. Game of coding with an unknown adversary. In2025 IEEE International Symposium on Information Theory (ISIT), Ann Arbor, MI, USA, 2025

    work page 2025

  32. [32]

    Game of coding: Coding theory in the presence of rational adversaries, motivated by decentralized machine learning,

    Hanzaleh Akbari Nodehi, Viveck R Cadambe, and Mohammad Ali Maddah-Ali. Game of coding: Coding theory in the presence of rational adversaries, motivated by decentralized machine learning.arXiv preprint arXiv:2601.02313, 2026

    work page arXiv 2026

  33. [33]

    Springer Science & Business Media, 2010

    Heinrich V on Stackelberg.Market structure and equilibrium. Springer Science & Business Media, 2010

    work page 2010

  34. [34]

    Adam: A method for stochastic optimization

    Diederik P Kingma and Jimmy Ba. Adam: A method for stochastic optimization. InInterna- tional Conference on Learning Representations, 2015

    work page 2015

  35. [35]

    Adadelta: an adaptive learning rate method.arXiv preprint arXiv:1212.5701,

    Matthew D Zeiler. Adadelta: an adaptive learning rate method.arXiv preprint arXiv:1212.5701, 2012

    work page arXiv 2012

  36. [36]

    Adaptive gradient methods with dynamic bound of learning rate,

    Liangchen Luo, Yuanhao Xiong, Yan Liu, and Xu Sun. Adaptive gradient methods with dynamic bound of learning rate.arXiv preprint arXiv:1902.09843, 2019

    work page arXiv 1902

  37. [37]

    Adaptive methods for nonconvex optimization.Advances in Neural Information Processing Systems, 31, 2018

    Manzil Zaheer, Sashank Reddi, Devendra Sachan, Satyen Kale, and Sanjiv Kumar. Adaptive methods for nonconvex optimization.Advances in Neural Information Processing Systems, 31, 2018

    work page 2018

  38. [38]

    Decoupled Weight Decay Regularization

    Ilya Loshchilov and Frank Hutter. Decoupled weight decay regularization.arXiv preprint arXiv:1711.05101, 2017

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  39. [39]

    Exponential moving average of weights in deep learning: Dynamics and benefits,

    Daniel Morales-Brotons, Thijs V ogels, and Hadrien Hendrikx. Exponential moving average of weights in deep learning: Dynamics and benefits.arXiv preprint arXiv:2411.18704, 2024. 12

    work page arXiv 2024

  40. [40]

    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

    work page 1964

  41. [41]

    Adafactor: Adaptive learning rates with sublinear memory cost

    Noam Shazeer and Mitchell Stern. Adafactor: Adaptive learning rates with sublinear memory cost. InInternational Conference on Machine Learning, pages 4596–4604. PMLR, 2018

    work page 2018

  42. [42]

    Momentum-based variance reduction in non-convex SGD.Advances in Neural Information Processing Systems, 32, 2019

    Ashok Cutkosky and Francesco Orabona. Momentum-based variance reduction in non-convex SGD.Advances in Neural Information Processing Systems, 32, 2019

    work page 2019

  43. [43]

    A stochastic approximation method.The Annals of Mathematical Statistics, pages 400–407, 1951

    Herbert Robbins and Sutton Monro. A stochastic approximation method.The Annals of Mathematical Statistics, pages 400–407, 1951

    work page 1951

  44. [44]

    Optimization methods for large-scale machine learning.SIAM Review, 60(2):223–311, 2018

    Léon Bottou, Frank E Curtis, and Jorge Nocedal. Optimization methods for large-scale machine learning.SIAM Review, 60(2):223–311, 2018

    work page 2018

  45. [45]

    Adaptive subgradient methods for online learning and stochastic optimization.Journal of Machine Learning Research, 12(7), 2011

    John Duchi, Elad Hazan, and Yoram Singer. Adaptive subgradient methods for online learning and stochastic optimization.Journal of Machine Learning Research, 12(7), 2011

    work page 2011

  46. [46]

    Adabelief optimizer: Adapting stepsizes by the belief in observed gradients.Advances in Neural Information Processing Systems, 33:18795–18806, 2020

    Juntang Zhuang, Tommy Tang, Yifan Ding, Sekhar C Tatikonda, Nicha Dvornek, Xenophon Papademetris, and James Duncan. Adabelief optimizer: Adapting stepsizes by the belief in observed gradients.Advances in Neural Information Processing Systems, 33:18795–18806, 2020

    work page 2020

  47. [47]

    On the Convergence of Adam and Beyond

    Sashank J Reddi, Satyen Kale, and Sanjiv Kumar. On the convergence of adam and beyond. arXiv preprint arXiv:1904.09237, 2019

    work page Pith review arXiv 1904

  48. [48]

    Adagrad stepsizes: Sharp convergence over nonconvex landscapes.Journal of Machine Learning Research, 21(219):1–30, 2020

    Rachel Ward, Xiaoxia Wu, and Leon Bottou. Adagrad stepsizes: Sharp convergence over nonconvex landscapes.Journal of Machine Learning Research, 21(219):1–30, 2020

    work page 2020

  49. [49]

    On the importance of initial- ization and momentum in deep learning

    Ilya Sutskever, James Martens, George Dahl, and Geoffrey Hinton. On the importance of initial- ization and momentum in deep learning. InInternational Conference on Machine Learning, pages 1139–1147. PMLR, 2013

    work page 2013

  50. [50]

    and Gu, Q

    Jinghui Chen, Dongruo Zhou, Yiqi Tang, Ziyan Yang, Yuan Cao, and Quanquan Gu. Closing the generalization gap of adaptive gradient methods in training deep neural networks.arXiv preprint arXiv:1806.06763, 2018

    work page arXiv 2018

  51. [51]

    Learning-rate-free learning by D-Adaptation

    Aaron Defazio and Konstantin Mishchenko. Learning-rate-free learning by D-Adaptation. In International Conference on Machine Learning, pages 7449–7479. PMLR, 2023

    work page 2023

  52. [52]

    Learning in stackelberg games with non-myopic agents

    Nika Haghtalab, Thodoris Lykouris, Sloan Nietert, and Alexander Wei. Learning in stackelberg games with non-myopic agents. InProceedings of the 23rd ACM Conference on Economics and Computation, pages 917–918, 2022

    work page 2022

  53. [53]

    Repeated contracting with multiple non-myopic agents: Policy regret and limited liability.arXiv preprint arXiv:2402.17108, 2024

    Natalie Collina, Varun Gupta, and Aaron Roth. Repeated contracting with multiple non-myopic agents: Policy regret and limited liability.arXiv preprint arXiv:2402.17108, 2024

    work page arXiv 2024

  54. [54]

    Pareto-optimal algo- rithms for learning in games

    Eshwar Ram Arunachaleswaran, Natalie Collina, and Jon Schneider. Pareto-optimal algo- rithms for learning in games. InProceedings of the 25th ACM Conference on Economics and Computation, 2024

    work page 2024

  55. [55]

    Levine, and Wolfgang Pesendorfer

    Marco Celentani, Drew Fudenberg, David K. Levine, and Wolfgang Pesendorfer. Maintaining a reputation against a long-lived opponent.Econometrica, 64(3):691–704, 1996

    work page 1996

  56. [56]

    Learning to manipulate a commitment optimizer.arXiv preprint arXiv:2302.11829, 2023

    Yurong Chen, Xiaotie Deng, Jiarui Gan, and Yuhao Li. Learning to manipulate a commitment optimizer.arXiv preprint arXiv:2302.11829, 2023

    work page arXiv 2023

  57. [57]

    Springer, Boston, MA, 2004

    Yurii Nesterov.Introductory Lectures on Convex Optimization: A Basic Course, volume 87 ofApplied Optimization. Springer, Boston, MA, 2004. ISBN 978-1-4020-7553-7. doi: 10.1007/978-1-4419-8853-9

    work page doi:10.1007/978-1-4419-8853-9 2004

  58. [58]

    arXiv preprint arXiv:1308.6370 , year=

    Mark Schmidt and Nicolas Le Roux. Fast convergence of stochastic gradient descent under a strong growth condition.arXiv, 2013. doi: 10.48550/arxiv.1308.6370. 13

    work page doi:10.48550/arxiv.1308.6370 2013

  59. [59]

    Gradient-based learning applied to document recognition.Proceedings of the IEEE, 86(11):2278–2324, 2002

    Yann LeCun, Léon Bottou, Yoshua Bengio, and Patrick Haffner. Gradient-based learning applied to document recognition.Proceedings of the IEEE, 86(11):2278–2324, 2002

    work page 2002

  60. [60]

    MNIST handwritten digit database.ATT Labs [Online], 2, 2010

    Yann LeCun, Corinna Cortes, and CJ Burges. MNIST handwritten digit database.ATT Labs [Online], 2, 2010. URLhttp://yann.lecun.com/exdb/mnist

    work page 2010

  61. [61]

    Deep residual learning for im- age recognition

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for im- age recognition. InProceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 770–778, 2016

    work page 2016

  62. [62]

    Learning multiple layers of features from tiny images

    Alex Krizhevsky. Learning multiple layers of features from tiny images. Technical report, University of Toronto, 2009

    work page 2009

  63. [63]

    Brendan McMahan, Eider Moore, Daniel Ramage, Seth Hampson, and Blaise Agüera y Arcas

    H. Brendan McMahan, Eider Moore, Daniel Ramage, Seth Hampson, and Blaise Agüera y Arcas. Communication-efficient learning of deep networks from decentralized data. InProceedings of the 20th International Conference on Artificial Intelligence and Statistics, pages 1273–1282. PMLR, 2017

    work page 2017

  64. [64]

    FLTrust: Byzantine-robust fed- erated learning via trust bootstrapping

    Xiaoyu Cao, Minghong Fang, Jia Liu, and Neil Zhenqiang Gong. FLTrust: Byzantine-robust fed- erated learning via trust bootstrapping. InNetwork and Distributed System Security Symposium, 2021

    work page 2021

  65. [65]

    FLAME: Taming backdoors in federated learning

    Thien Duc Nguyen, Phillip Rieger, Huili Chen, Hossein Yalame, Helen Möllering, Hossein Fereidooni, Samuel Marchal, Markus Miettinen, Azalia Mirhoseini, Sadegh Zeitouni, et al. FLAME: Taming backdoors in federated learning. In31st USENIX Security Symposium, pages 1415–1432, 2022

    work page 2022

  66. [66]

    Fed- DMC: Efficient and robust federated learning via detecting malicious clients.IEEE Transactions on Dependable and Secure Computing, 21(6):5259–5274, 2024

    Xutong Mu, Ke Cheng, Yuxin Shen, Xin Li, Zhen Chang, Tao Zhang, and Xiaosong Ma. Fed- DMC: Efficient and robust federated learning via detecting malicious clients.IEEE Transactions on Dependable and Secure Computing, 21(6):5259–5274, 2024

    work page 2024

  67. [67]

    FedID: Enhancing federated learning security through dynamic identification.IEEE Transactions on Pattern Analysis and Machine Intelligence, 47(10):8907–8922, 2025

    Siquan Huang, Yijiang Li, Chong Chen, Ying Gao, and Xiping Hu. FedID: Enhancing federated learning security through dynamic identification.IEEE Transactions on Pattern Analysis and Machine Intelligence, 47(10):8907–8922, 2025

    work page 2025

  68. [68]

    RepuNet: A reputation system for mitigating malicious clients in decentralized federated learning.arXiv preprint arXiv:2506.19892, 2025

    Isaac Marroqui Penalva, Enrique Tomás Martínez Beltrán, Manuel Gil Pérez, and Alberto Huer- tas Celdrán. RepuNet: A reputation system for mitigating malicious clients in decentralized federated learning.arXiv preprint arXiv:2506.19892, 2025

    work page arXiv 2025

  69. [69]

    FLARE: Adaptive Multi-Dimensional Reputation for Robust Client Reliability in Federated Learning,

    Abolfazl Younesi, Leon Kiss, Zahra Najafabadi Samani, Juan Aznar Poveda, and Thomas Fahringer. FLARE: Adaptive multi-dimensional reputation for robust client reliability in federated learning.arXiv preprint arXiv:2511.14715, 2025

    work page internal anchor Pith review arXiv 2025

  70. [70]

    Bitcoin: A peer-to-peer electronic cash system, 2008

    Satoshi Nakamoto. Bitcoin: A peer-to-peer electronic cash system, 2008

    work page 2008

  71. [71]

    Ethereum white paper.GitHub repository, 1:22–23, 2013

    Vitalik Buterin et al. Ethereum white paper.GitHub repository, 1:22–23, 2013

    work page 2013

  72. [72]

    Proofs, arguments, and zero-knowledge.Foundations and Trends® in Privacy and Security, 4(2–4):117–660, 2022

    Justin Thaler. Proofs, arguments, and zero-knowledge.Foundations and Trends® in Privacy and Security, 4(2–4):117–660, 2022

    work page 2022

  73. [73]

    ZEN: An optimizing compiler for verifiable, zero-knowledge neural network inferences.Cryptology ePrint Archive, 2021

    Boyuan Feng, Lianke Qin, Zhenfei Zhang, Yufei Ding, and Shumo Chu. ZEN: An optimizing compiler for verifiable, zero-knowledge neural network inferences.Cryptology ePrint Archive, 2021

    work page 2021

  74. [74]

    ZkCNN: Zero-knowledge proofs for convolutional neural network predictions and accuracy

    Tianyi Liu, Xiang Xie, and Yupeng Zhang. ZkCNN: Zero-knowledge proofs for convolutional neural network predictions and accuracy. InProceedings of the 2021 ACM SIGSAC Conference on Computer and Communications Security, pages 2968–2985, 2021

    work page 2021

  75. [75]

    Zero-knowledge proof meets machine learning in verifiability: A survey.arXiv preprint arXiv:2310.14848, 2023

    Zhibo Xing, Zijian Zhang, Jiamou Liu, Ziang Zhang, Meng Li, Liehuang Zhu, and Giovanni Russello. Zero-knowledge proof meets machine learning in verifiability: A survey.arXiv preprint arXiv:2310.14848, 2023

    work page arXiv 2023

  76. [76]

    SecureML: A system for scalable privacy-preserving machine learning

    Payman Mohassel and Yupeng Zhang. SecureML: A system for scalable privacy-preserving machine learning. In2017 IEEE Symposium on Security and Privacy (SP), pages 19–38. IEEE, 2017. 14

    work page 2017

  77. [77]

    vCNN: Verifiable convolutional neu- ral network based on zk-SNARKs.IEEE Transactions on Dependable and Secure Computing, 2024

    Seunghwa Lee, Hankyung Ko, Jihye Kim, and Hyunok Oh. vCNN: Verifiable convolutional neu- ral network based on zk-SNARKs.IEEE Transactions on Dependable and Secure Computing, 2024

    work page 2024

  78. [78]

    Mystique: Efficient conversions for Zero-Knowledge proofs with applications to machine learning

    Chenkai Weng, Kang Yang, Xiang Xie, Jonathan Katz, and Xiao Wang. Mystique: Efficient conversions for Zero-Knowledge proofs with applications to machine learning. In30th USENIX Security Symposium (USENIX Security 21), pages 501–518, 2021

    work page 2021

  79. [79]

    Interactive proofs for rounding arithmetic.IEEE Access, 10:122706–122725, 2022

    Shuo Chen, Jung Hee Cheon, Dongwoo Kim, and Daejun Park. Interactive proofs for rounding arithmetic.IEEE Access, 10:122706–122725, 2022

    work page 2022

  80. [80]

    Succinct zero-knowledge for floating point computations

    Sanjam Garg, Abhishek Jain, Zhengzhong Jin, and Yinuo Zhang. Succinct zero-knowledge for floating point computations. InProceedings of the 2022 ACM SIGSAC Conference on Computer and Communications Security, pages 1203–1216, 2022

    work page 2022

Showing first 80 references.