Pith. sign in

REVIEW 3 major objections 4 minor 42 references

TACO: Tackling Over-correction in Federated Learning with Tailored Adaptive Correction

T0 review · 3 major / 4 minor · reviewed 2026-08-16 · deepseek-v4-flash

Pith's one-line read TACO claims that per-client correction coefficients eliminate the over-correction caused by uniform coefficients in federated learning, and supports this with a convergence analysis and experiments on eight datasets.

desk verdict Solid empirical study of a lightweight adaptive-correction FL method, but the advertised convergence analysis is invalid and the optimality corollary is arithmetically wrong. read the letter →

arxiv 2504.17528 v1 pith:Q62G5UFL submitted 2025-04-24 cs.LG cs.AI

classification cs.LGcs.AI
keywords federatedlearningnon-IIDdataover-correctionclientdriftclient-specificcorrectionconvergenceanalysistime-to-accuracyfreeloaderdetection
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper tries to establish that a hidden 'over-correction' effect, not the absence of correction, explains why many federated-learning fixes for non-IID data (FedProx, Scaffold, STEM, FedACG) often do worse than plain FedAvg. The proposed remedy, TACO, gives each client its own correction coefficient, computed from the size and direction of that client's local gradient relative to the global gradient, and uses the same coefficient as an aggregation weight. Because clients with more skewed local data are corrected harder and weighted less, the paper argues, the over-correction introduced by uniform coefficients disappears. This would matter because non-IID data is the standard obstacle in federated learning, and the paper's design adds little client-side computation, so better accuracy could come without sacrificing training speed.

What carries the argument

The central object is the client-specific correction coefficient $\alpha_i^t$ in Eq. (7), built from the magnitude of the client's accumulated local gradient $\|\Delta_i^{t-1}\|$ divided by the sum of all clients' magnitudes, multiplied by the clipped cosine similarity between $\Delta_i^{t-1}$ and the average global gradient. It enters the local update $w_{i,k+1}^t = w_{i,k}^t - \eta_l(g_{i,k}^t + \gamma(1-\alpha_i^t)\Delta^t)$ and the server aggregation $\Delta^{t+1} = \frac{1}{K\eta_l}\frac{\sum_{i\in\mathcal{N}}\alpha_i^{t+1}\Delta_i^t}{\sum_{i\in\mathcal{N}}\alpha_i^{t+1}}$. The same coefficient thus simultaneously decides how strongly a client's local update is pulled toward the global model and how much weight that client's gradient gets in aggregation, which is the mechanism that turns the paper's over-correction diagnosis into a trainable algorithm.

What would settle it

Check whether $E\|\Delta_i^{t-1}\| \le (\mu_i/c_i)\|\nabla f(w^{t-1})\|$ holds on a two-client example that satisfies Assumption 2: take both client directions close to the global gradient but give client 1 a much larger local-gradient magnitude than client 2; if the bound fails while the stated cosine-similarity assumption holds, the $Y_t$ term that Corollary 2 minimizes is an artifact of the proof rather than a real over-correction effect. An experimental counterpart is to compare TACO with FedProx whose $\zeta_i$ is tuned per client; matching accuracy would indicate the uniform coefficient, not the correction direction, is what matters.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that the error bound for federated optimization contains a correction-induced term $Y_t = \frac{L^2G^2}{K^2N^4\eta_l^2}\big(\sum_{i\in\mathcal{N}}(1-\alpha_i^t)\sum_{i\in\mathcal{N}}\frac{\mu_i}{c_i}\big)^2$ that grows when all clients share one correction coefficient. TACO chooses $\alpha_i^t$ from $\|\Delta_i^{t-1}\|$ and its cosine similarity to the global gradient, and Corollary 2 states that the optimal choice sets $(1-\alpha_i^t)$ proportional to $\mu_i/c_i$, which minimizes $Y_t$. The paper claims this is the first convergence analysis to pinpoint uniform correction coefficients as the root cause of over-correction, and it reports that the same $\alpha_i^t$ doubles as a freeloader detector because clients that merely re-upload the previous global gradient receive unusually high $\alpha$ values. Empirically, TACO is reported to exceed all six baselines on eight datasets in round-to-accuracy and to reach target accuracy with 25.6–62.7% less client computation time than FedAvg.

Load-bearing premise

The proof's main conclusion depends on assuming that the expected size of a client's local gradient is no more than a fixed multiple of the size of the true global gradient, yet the assumptions stated in the paper only constrain the direction of the expected gradient, not its size; the over-correction term in the error bound comes from exactly this unsupported step.

Editorial extensions

If this is right

  • Uniform-coefficient correction methods can be read as a single flawed family: FedProx, Scaffold, STEM, and FedACG all enlarge the error term $Y_t$ when their shared coefficient does not match each client's skew, which the paper argues explains their instability and convergence failures on SVHN and CIFAR-10.
  • Replacing a baseline's uniform coefficient with TACO's tailored coefficients yields the paper's reported gains (FedProx on SVHN and Scaffold on CIFAR-10), suggesting the correction coefficient, not the surrounding method, carries much of the non-IID benefit.
  • TACO reaches target accuracy in fewer communication rounds than every baseline on the tested datasets, and cuts client-side computation time by 25.6–62.7% relative to FedAvg.
  • The convergence rate becomes $O\big(\sqrt{L/T} + \sqrt[3]{Y/T^2}\big)$, and when all $\alpha_i^t$ are equal it reduces to the existing FedCM rate, so the analysis nests prior uniform-coefficient results as a special case.
  • Freeloaders that re-upload the previous global gradient receive high $\alpha_i^t$ and can be expelled with threshold $\kappa$ over $\lambda$ rounds, with the reported operating region $\kappa\in[0.6,0.8]$, $\lambda=T/5$ giving 100% true positives and 0% false positives on FMNIST.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Beyond the paper's own claims, a testable extension is that other drift-correction families (FedDyn, FedDC, FedNova) should also improve when their uniform regularizer or weighting is replaced by a per-client $\alpha$-style term, since the paper only demonstrates the swap for FedProx and Scaffold.
  • The paper's time-to-accuracy re-evaluation predicts that communication-round comparisons favor algorithms with heavy local computation; an independent benchmark reporting both metrics could reveal how much of the reported TACO advantage is speed rather than accuracy.
  • If the over-correction diagnosis is right, tuning a uniform coefficient more carefully does not remove the problem; it only picks which clients to overcorrect. A direct check would compare TACO against FedProx with per-client $\zeta_i$ chosen by grid search, since matching performance would suggest the uniform-coefficient story is not the whole explanation.
  • The freeloader test relies on lazy clients behaving predictably (re-uploading the previous global gradient), so a plausible evasion is to copy a randomly selected honest client's gradient; whether TACO's $\alpha$ signature still exposes that variant is an open empirical question.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 4 minor

Summary. The paper proposes TACO, a federated learning algorithm that assigns per-client, time-varying correction coefficients α_i^t, used both in local updates and in server-side aggregation. The authors claim to identify a previously hidden 'over-correction' phenomenon caused by uniform correction coefficients in existing FL methods, and they support this with a convergence analysis (Theorem 1 and Corollaries 1–2) as well as experiments across eight datasets and several baselines. The central theoretical claim is that the error term Y_t in the bound is minimized when (1−α_i^t) ∝ μ_i/c_i, which motivates the particular form of α_i^t in Eq. (7).

Significance. If the theoretical results were sound, TACO would be a useful lightweight contribution to non-IID federated learning: it requires no auxiliary parameters, adds little per-round computation, and the time-to-accuracy comparison is a valuable evaluation lens. The experimental section is broad, comparing six baselines across eight datasets and including ablation, scalability, sensitivity, and freeloader-detection studies. However, the load-bearing theoretical arguments are not valid: a key inequality in the appendix is unjustified, the main recursion in Lemma 1 is inconsistent with the aggregation rule, and the optimality statement in Corollary 2 does not follow from the displayed bound. The empirical breadth does not compensate for the unsupported central claims, since the paper's novelty is advertised as the first convergence analysis that reveals the root cause of over-correction.

major comments (3)
  1. [Appendix, proof of Theorem 1] The step 'Using Eq. (11)/Eq. (12), we have E||Δ_i^{t−1}|| ≤ μ_i/c_i ||∇f(w^{t−1})||' is invalid. Assumption 2 bounds the inner product and cosine similarity involving the expectation E[Δ_i^{t−1}]; it yields a bound on ||E[Δ_i^{t−1}]||, not on E||Δ_i^{t−1}||. Jensen's inequality gives the reverse direction, E||Δ_i^{t−1}|| ≥ ||E[Δ_i^{t−1}]||, and the expectation of the norm can be much larger. This step is load-bearing because it is the point where the term Y_t is introduced in Theorem 1, and Corollary 1 inherits this error.
  2. [Corollary 2] The claimed optimization is not correct. In the expression Y_t = (L^2 G^2)/(K^2 N^4 η_l^2) (Σ_i (1−α_i^t))^2 (Σ_i μ_i/c_i)^2, the second factor is independent of the correction coefficients. Minimizing over (1−α_i^t) subject to Σ_i (1−α_i^t) ≥ σ only forces the sum to σ; any allocation with that sum gives the same Y_t. Thus the statement that (1−α_i^t) ∝ μ_i/c_i is not derived from the bound, and the claimed optimality of the tailored coefficients in Eq. (7) is unsupported.
  3. [Lemma 1 and Eq. (9)] Lemma 1 (Δ^{t+1} = tildeΔ^t + (1−α^t)Δ^t) does not follow from the aggregation rule in Eq. (9) when the coefficients α_i^{t+1} differ across clients. Expanding Δ_i^t from Eq. (8) with γ = 1 gives terms of the form α_i^{t+1}(1−α_i^t)Δ^t inside the summation, and the outer factor (Σ_i α_i^{t+1})/N in Eq. (9) does not reduce this to the average coefficient α^t = (1/N)Σ_i α_i^t. Consequently Lemma 2 and the subsequent bound based on z^{t+1} = z^t − η_g tildeΔ^t are not justified.
minor comments (4)
  1. [Section V-B] The claimed improvement range '2.76%–58.68%' is not clearly defined; please specify the comparison basis (final accuracy after fixed rounds, or rounds to target accuracy).
  2. [Fig. 5] The x-axis labels in Fig. 5(a) and 5(b) are compressed ('FedAvg/FGFedProxScaffold'), making the figure difficult to read.
  3. [Section IV-A] The timing of the correction coefficients should be stated explicitly: Eq. (7) computes α_i^t using Δ_i^{t−1} and the previous round's average gradient, while Eq. (9) uses α_i^{t+1} computed after receiving Δ_i^t; the relationship to the local update in Eq. (8) is not immediately transparent.
  4. [Appendix, proof of Theorem 1] The proof sets γ = 1 'for brevity', but γ is a tunable hyperparameter in the experiments; a sentence explaining why the analysis extends to general γ would help.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity found: TACO's theoretical gaps are mathematical errors, not self-referential reductions.

full rationale

I walked the derivation chain from Assumptions 1-3 through Theorem 1 and Corollaries 1-2. The central claim that tailored coefficients are optimal is not forced by construction: the displayed error term Y_t is a product of (sum_i(1-alpha_i))^2 and (sum_i mu_i/c_i)^2, so minimizing Y_t over alpha_i with a sum constraint does not determine per-client proportions; the Cauchy-Schwarz step in Corollary 2 is an unjustified assertion, not a conclusion already contained in the input. Likewise, the appendix line 'Using Eq. (11)/Eq. (12), we have E||Delta_i|| <= mu_i/c_i ||grad f||' misapplies Assumption 2, which bounds the norm of an expectation rather than the expectation of a norm; but this is an invalid inference, not an equation that is equivalent to its own input by definition. Lemma 1's replacement of the tailored aggregation by a single average coefficient is inconsistent rather than circular. The empirical evaluation is independent: it compares TACO against six baselines on external datasets, and the only author overlap in the references is a generic citation for standard assumptions, which is not load-bearing. The paper's proof has serious correctness risks, but under the circularity rubric there is no demonstrated reduction of a prediction to a fitted input or a self-citation chain, so no significant circularity is found.

Assumptions & free parameters 3 free parameters · 5 assumptions · 0 invented entities

The central claim rests on standard smoothness and bounded-gradient assumptions, plus two non-standard assumptions introduced to make the proof tractable: Assumption 2's per-client bounds and Lemmas 1-2 that define the recursion used in the bound. The per-client bounds are not verified empirically, and the lemmas are inconsistent with the algorithm equations as written. Free parameters gamma, kappa, and lambda are tuned empirically on the evaluation datasets.

free parameters (3)
  • gamma (maximum correction factor) = gamma = 1/K default; sensitivity sweep over {0, 0.001, 0.01, 0.1, 1.0}
    Controls the strength of the correction term in Eq. (8). The optimal value is found empirically per dataset and depends on K, so it is fitted to data.
  • kappa (freeloader detection threshold) = 0.6 default (swept 0.4 to 1.0)
    Threshold in Eq. (10) to flag suspicious clients; chosen based on the sensitivity table to balance TPR/FPR on FMNIST.
  • lambda (eviction strikes) = T/5 default (swept T/10 to T/2)
    Number of times a client must be flagged before expulsion; tuned per dataset.
assumptions (5)
  • standard math Assumption 1: L-smooth local losses
    Common smoothness condition in FL convergence analysis.
  • standard math Assumption 3: bounded true gradient norm G
    Standard bounded gradient assumption.
  • ad hoc to paper Assumption 2: per-client cosine similarity and magnitude bounds (mu_i, c_i) on the expected accumulated gradient
    Introduced specifically to quantify client heterogeneity in a way that makes the proof's error term Yt depend on (1-alpha_i) and mu_i/c_i. No empirical verification is provided for these bounds, and the proof uses a stronger inequality (expectation of norm) than the assumption provides.
  • ad hoc to paper Lemma 1: Delta^{t+1} = tildeDelta^t + (1-alpha^t)Delta^t
    This update rule is asserted with a proof sketch but does not follow from Eq. (9) when alpha_i differ across clients; the proof of Theorem 1 relies on it.
  • ad hoc to paper Lemma 2: z^{t+1} = z^t - eta_g tildeDelta^t
    Asserted without a detailed derivation; this exact linear reduction to the average client gradient ignores the correction and tailored aggregation terms in TACO, yet it drives the main loss descent argument.

how reviews work

0 comments
Cite this review

Pith. "Pith review of TACO: Tackling Over-correction in Federated Learning with Tailored Adaptive Correction." pith.science (2026). https://pith.science/paper/Q62G5UFL

@misc{pith2026250417528,
  author       = {Pith},
  title        = {Pith review of: TACO: Tackling Over-correction in Federated Learning with Tailored Adaptive Correction},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/Q62G5UFL}},
  note         = {Machine review of arXiv:2504.17528}
}
read the original abstract

Non-independent and identically distributed (Non-IID) data across edge clients have long posed significant challenges to federated learning (FL) training in edge computing environments. Prior works have proposed various methods to mitigate this statistical heterogeneity. While these works can achieve good theoretical performance, in this work we provide the first investigation into a hidden over-correction phenomenon brought by the uniform model correction coefficients across clients adopted by existing methods. Such over-correction could degrade model performance and even cause failures in model convergence. To address this, we propose TACO, a novel algorithm that addresses the non-IID nature of clients' data by implementing fine-grained, client-specific gradient correction and model aggregation, steering local models towards a more accurate global optimum. Moreover, we verify that leading FL algorithms generally have better model accuracy in terms of communication rounds rather than wall-clock time, resulting from their extra computation overhead imposed on clients. To enhance the training efficiency, TACO deploys a lightweight model correction and tailored aggregation approach that requires minimum computation overhead and no extra information beyond the synchronized model parameters. To validate TACO's effectiveness, we present the first FL convergence analysis that reveals the root cause of over-correction. Extensive experiments across various datasets confirm TACO's superior and stable performance in practice.

Figures

Figures reproduced from arXiv: 2504.17528 by the authors.

Figure 1
Figure 1. Federated Learning with non-IID data: Uniform cor [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 3
Figure 3. Clients with larger θi and magnitude of the local gradient ∆t i need larger correction factors 1 − α t i . and thus longer wall-clock time per round. Figs 2c and 2d together with Table I show that most algorithms, especially STEM, significantly increase the computation time of local up￾dates in every FL round, which leads to their mediocre time-to￾accuracy performance. We also observe that the incorporation of regul… view at source ↗
Figure 2
Figure 2. Round-to-accuracy and time-to-accuracy re-evaluations [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figures from the paper (4 more)
Figure 3
Figure 3. Figure 3: Fig. 3-Left shows that client 1 should have a larger [PITH_FULL_IMAGE:figures/full_fig_p005_3.png]
Figure 4
Figure 4. Figure 4: Cumulative local training time required by different algorithms to achieve the target accuracy. [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]
Figure 6
Figure 6. Figure 6: Performance gain in prior methods using TACO. [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]
Figure 7
Figure 7. Figure 7: Sensitivity of γ. {T /10, T /5, T /2}, where T is the number of communication rounds. As illustrated in Table VIII, an increase in the values of κ and λ represents a stricter criterion to identify a client as a freeloader, which reduces the probability of falsely cate￾…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

42 extracted references · 17 canonical work pages

  1. [11]

    Federated learning on non-iid data silos: An experimental study,

    Q. Li, Y . Diao, Q. Chen, and B. He, “Federated learning on non-iid data silos: An experimental study,” in 2022 IEEE 38th International Conference on Data Engineering (ICDE) . IEEE, 2022, pp. 965–978

  2. [1]

    Edge intelligence: Paving the last mile of artificial intelligence with edge computing,

    Z. Zhou, X. Chen, E. Li, L. Zeng, K. Luo, and J. Zhang, “Edge intelligence: Paving the last mile of artificial intelligence with edge computing,” Proceedings of the IEEE , vol. 107, no. 8, pp. 1738–1762, 2019

  3. [2]

    Communication-efficient learning of deep networks from decentralized data,

    B. McMahan, E. Moore, D. Ramage, S. Hampson, and B. A. y Arcas, “Communication-efficient learning of deep networks from decentralized data,” in Proc. of Artificial intelligence and statistics , 2017

  4. [3]

    Accelerating federated learning with data and model parallelism in edge computing,

    Y . Liao, Y . Xu, H. Xu, Z. Yao, L. Wang, and C. Qiao, “Accelerating federated learning with data and model parallelism in edge computing,” IEEE/ACM Transactions on Networking , 2023

  5. [4]

    Federated learning over wireless networks: Convergence analysis and resource allocation,

    C. T. Dinh, N. H. Tran, M. N. Nguyen, C. S. Hong, W. Bao, A. Y . Zomaya, and V . Gramoli, “Federated learning over wireless networks: Convergence analysis and resource allocation,” IEEE/ACM Transactions on Networking, vol. 29, no. 1, pp. 398–409, 2020

  6. [5]

    Accelerating and securing federated learning with stateless in-network aggregation at the edge,

    J. Xia, W. Wu, L. Luo, G. Cheng, D. Guo, and Q. Nian, “Accelerating and securing federated learning with stateless in-network aggregation at the edge,” in 2024 IEEE 44th International Conference on Distributed Computing Systems (ICDCS) . IEEE, 2024, pp. 692–702

  7. [6]

    Federated optimization in heterogeneous networks,

    T. Li, A. K. Sahu, M. Zaheer, M. Sanjabi, A. Talwalkar, and V . Smith, “Federated optimization in heterogeneous networks,” Proceedings of Machine learning and systems , vol. 2, pp. 429–450, 2020

  8. [7]

    The limitations of federated learning in sybil settings,

    C. Fung, C. J. Yoon, and I. Beschastnikh, “The limitations of federated learning in sybil settings,” in 23rd International Symposium on Research in Attacks, Intrusions and Defenses (RAID 2020) , 2020, pp. 301–316

Show all 42 references
  1. [8]

    Scaffold: Stochastic controlled averaging for federated learn- ing,

    S. P. Karimireddy, S. Kale, M. Mohri, S. Reddi, S. Stich, and A. T. Suresh, “Scaffold: Stochastic controlled averaging for federated learn- ing,” in International conference on machine learning . PMLR, 2020, pp. 5132–5143

  2. [9]

    Stem: A stochastic two-sided momentum algorithm achieving near-optimal sample and communication complexities for federated learning,

    P. Khanduri, P. Sharma, H. Yang, M. Hong, J. Liu, K. Rajawat, and P. Varshney, “Stem: A stochastic two-sided momentum algorithm achieving near-optimal sample and communication complexities for federated learning,” Advances in Neural Information Processing Systems, vol. 34, pp....

  3. [10]

    Communication-efficient federated learn- ing with accelerated client gradient,

    G. Kim, J. Kim, and B. Han, “Communication-efficient federated learn- ing with accelerated client gradient,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2024, pp. 12 385–12 394

  4. [12]

    Tackling the objective inconsistency problem in heterogeneous federated optimiza- tion,

    J. Wang, Q. Liu, H. Liang, G. Joshi, and H. V . Poor, “Tackling the objective inconsistency problem in heterogeneous federated optimiza- tion,” Advances in neural information processing systems , vol. 33, pp. 7611–7623, 2020

  5. [13]

    To talk or to work: Dynamic batch sizes assisted time efficient federated learn- ing over future mobile edge devices,

    D. Shi, L. Li, M. Wu, M. Shu, R. Yu, M. Pan, and Z. Han, “To talk or to work: Dynamic batch sizes assisted time efficient federated learn- ing over future mobile edge devices,” IEEE Transactions on Wireless Communications, vol. 21, no. 12, pp. 11 038–11 050, 2022

  6. [14]

    Reading digits in natural images with unsupervised feature learning,

    Y . Netzer, T. Wang, A. Coates, A. Bissacco, B. Wu, and A. Y . Ng, “Reading digits in natural images with unsupervised feature learning,” 2011

  7. [15]

    Free-rider attacks on model aggregation in federated learning,

    Y . Fraboni, R. Vidal, and M. Lorenzi, “Free-rider attacks on model aggregation in federated learning,” in International Conference on Artificial Intelligence and Statistics . PMLR, 2021, pp. 1846–1854

  8. [16]

    Enabling long-term cooperation in cross-silo federated learning: A repeated game perspective,

    N. Zhang, Q. Ma, and X. Chen, “Enabling long-term cooperation in cross-silo federated learning: A repeated game perspective,” IEEE Transactions on Mobile Computing, vol. 22, no. 7, pp. 3910–3924, 2022

  9. [17]

    Free-riders in federated learning: Attacks and defenses,

    J. Lin, M. Du, and J. Liu, “Free-riders in federated learning: Attacks and defenses,” arXiv preprint arXiv:1911.12560 , 2019

  10. [18]

    Toward free-riding attack on cross-silo federated learning through evolutionary game,

    T. Chen, F. Wang, W. Qiu, Q. Zhang, Z. Xiong, and Z. Zheng, “Toward free-riding attack on cross-silo federated learning through evolutionary game,” in 2024 IEEE 44th International Conference on Distributed Computing Systems (ICDCS) . IEEE, 2024, pp. 869–880

  11. [19]

    Fedcm: Federated learning with client-level momentum,

    J. Xu, S. Wang, L. Wang, and A. C.-C. Yao, “Fedcm: Federated learning with client-level momentum,” arXiv preprint arXiv:2106.10874 , 2021

  12. [20]

    Optimization methods for large- scale machine learning,

    L. Bottou, F. E. Curtis, and J. Nocedal, “Optimization methods for large- scale machine learning,” SIAM review, vol. 60, no. 2, pp. 223–311, 2018

  13. [21]

    Adaptive federated learning in resource constrained edge computing systems,

    S. Wang, T. Tuor, T. Salonidis, K. K. Leung, C. Makaya, T. He, and K. Chan, “Adaptive federated learning in resource constrained edge computing systems,” IEEE journal on selected areas in communications, vol. 37, no. 6, pp. 1205–1221, 2019

  14. [22]

    Feder- ated learning under heterogeneous and correlated client availability,

    A. Rodio, F. Faticanti, O. Marfoq, G. Neglia, and E. Leonardi, “Feder- ated learning under heterogeneous and correlated client availability,” in IEEE INFOCOM 2023-IEEE Conference on Computer Communications. IEEE, 2023, pp. 1–10

  15. [23]

    Federated learning with flexible control,

    S. Wang, J. Perazzone, M. Ji, and K. S. Chan, “Federated learning with flexible control,” in IEEE INFOCOM 2023-IEEE Conference on Computer Communications. IEEE, 2023, pp. 1–10

  16. [24]

    Dynamite: Dynamic interplay of mini-batch size and aggregation frequency for federated learning with static and streaming dataset,

    W. Liu, X. Zhang, J. Duan, C. Joe-Wong, Z. Zhou, and X. Chen, “Dynamite: Dynamic interplay of mini-batch size and aggregation frequency for federated learning with static and streaming dataset,” IEEE Transactions on Mobile Computing , 2023

  17. [25]

    Advances and open problems in federated learning,

    P. Kairouz, H. B. McMahan, B. Avent, A. Bellet, M. Bennis, A. N. Bhagoji, K. Bonawitz, Z. Charles, G. Cormode, R. Cummings et al. , “Advances and open problems in federated learning,” Foundations and Trends® in Machine Learning , vol. 14, no. 1–2, pp. 1–210, 2021

  18. [26]

    Fedmos: Taming client drift in federated learning with double momentum and adaptive selection,

    X. Wang, Y . Chen, Y . Li, X. Liao, H. Jin, and B. Li, “Fedmos: Taming client drift in federated learning with double momentum and adaptive selection,” in IEEE INFOCOM, 2023

  19. [27]

    On the convergence of fedavg on non-iid data,

    X. Li, K. Huang, W. Yang, S. Wang, and Z. Zhang, “On the convergence of fedavg on non-iid data,” arXiv preprint arXiv:1907.02189 , 2019

  20. [28]

    Becker and R

    B. Becker and R. Kohavi, “Adult,” UCI Machine Learning Repository, 1996, DOI: https://doi.org/10.24432/C5XW20

  21. [29]

    Deep residual learning for image recognition,

    K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” in Proceedings of the IEEE conference on computer vision and pattern recognition , 2016, pp. 770–778

  22. [30]

    Autofl: A bayesian game approach for autonomous client participation in federated edge learning,

    M. Hu, W. Yang, Z. Luo, X. Liu, Y . Zhou, X. Chen, and D. Wu, “Autofl: A bayesian game approach for autonomous client participation in federated edge learning,” IEEE Transactions on Mobile Computing , vol. 23, no. 1, pp. 194–208, 2022

  23. [31]

    Leaf: A benchmark for federated settings,

    S. Caldas, S. M. K. Duddu, P. Wu, T. Li, J. Kone ˇcn`y, H. B. McMahan, V . Smith, and A. Talwalkar, “Leaf: A benchmark for federated settings,” arXiv preprint arXiv:1812.01097 , 2018

  24. [32]

    A hierarchical knowledge transfer framework for heterogeneous federated learning,

    Y . Deng, J. Ren, C. Tang, F. Lyu, Y . Liu, and Y . Zhang, “A hierarchical knowledge transfer framework for heterogeneous federated learning,” in IEEE INFOCOM 2023-IEEE Conference on Computer Communications. IEEE, 2023, pp. 1–10

  25. [33]

    Distributionally robust federated learning for network traffic classification with noisy labels,

    S. Shi, Y . Guo, D. Wang, Y . Zhu, and Z. Han, “Distributionally robust federated learning for network traffic classification with noisy labels,” IEEE Transactions on Mobile Computing , 2023

  26. [34]

    Heterogeneity- aware federated learning with adaptive client selection and gradient compression,

    Z. Jiang, Y . Xu, H. Xu, Z. Wang, and C. Qian, “Heterogeneity- aware federated learning with adaptive client selection and gradient compression,” in IEEE INFOCOM 2023-IEEE Conference on Computer Communications. IEEE, 2023, pp. 1–10

  27. [35]

    Internal cross-layer gradients for extending homogeneity to heterogeneity in federated learning,

    Y .-H. Chan, R. Zhou, R. Zhao, Z. JIANG, and E. C. Ngai, “Internal cross-layer gradients for extending homogeneity to heterogeneity in federated learning,” in The Twelfth International Conference on Learning Representations, 2024

  28. [36]

    Can federated learning clients be lightweight? a plug-and-play symmetric conversion module,

    J. Liu, H. Huang, C. Wang, R. Li, T. Car, Q. Yang, and Z. Zheng, “Can federated learning clients be lightweight? a plug-and-play symmetric conversion module,” in 2024 IEEE 44th International Conference on Distributed Computing Systems (ICDCS) . IEEE, 2024, pp. 809–820

  29. [37]

    Federated learning based on dynamic regularization,

    D. A. E. Acar, Y . Zhao, R. M. Navarro, M. Mattina, P. N. Whatmough, and V . Saligrama, “Federated learning based on dynamic regularization,” arXiv preprint arXiv:2111.04263 , 2021

  30. [38]

    Feddc: Federated learning with non-iid data via local drift decoupling and correction,

    L. Gao, H. Fu, L. Li, Y . Chen, M. Xu, and C.-Z. Xu, “Feddc: Federated learning with non-iid data via local drift decoupling and correction,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2022, pp. 10 112–10 121

  31. [39]

    Towards flexible device participation in federated learning,

    Y . Ruan, X. Zhang, S.-C. Liang, and C. Joe-Wong, “Towards flexible device participation in federated learning,” in International Conference on Artificial Intelligence and Statistics . PMLR, 2021, pp. 3403–3411

  32. [40]

    Mime: Mimicking centralized stochastic algorithms in federated learning,

    S. P. Karimireddy, M. Jaggi, S. Kale, M. Mohri, S. J. Reddi, S. U. Stich, and A. T. Suresh, “Mime: Mimicking centralized stochastic algorithms in federated learning,” arXiv preprint arXiv:2008.03606 , 2020

  33. [41]

    Ma- chine learning with adversaries: Byzantine tolerant gradient descent,

    P. Blanchard, E. M. El Mhamdi, R. Guerraoui, and J. Stainer, “Ma- chine learning with adversaries: Byzantine tolerant gradient descent,” Advances in neural information processing systems , vol. 30, 2017

  34. [42]

    Feder- ated learning with compression: Unified analysis and sharp guarantees,

    F. Haddadpour, M. M. Kamani, A. Mokhtari, and M. Mahdavi, “Feder- ated learning with compression: Unified analysis and sharp guarantees,” in International Conference on Artificial Intelligence and Statistics . PMLR, 2021, pp. 2350–2358

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.