Pith. sign in

REVIEW 4 major objections 5 minor 32 references

Hierarchical Federated Learning Across Heterogeneous Cellular Networks

T0 review · 4 major / 5 minor · reviewed 2026-08-14 · deepseek-v4-flash

Pith's one-line read Hierarchical federated learning that routes updates through small cells and sends only sparse gradients can cut communication latency dramatically while keeping CIFAR-10 accuracy close to centralized training.

desk verdict A useful integration of hierarchical FL, gradient sparsification, and subcarrier allocation; the latency model is communication-only and the optimality proof is sketchy, but the design direction is sound. read the letter →

arxiv 1909.02362 v1 pith:B46BPFDR submitted 2019-09-05 cs.LG cs.DCcs.ITeess.SPmath.ITstat.ML

classification cs.LGcs.DCcs.ITeess.SPmath.ITstat.ML
keywords federatedlearninghierarchicalheterogeneouscellularnetworksgradientsparsificationcommunicationlatencyerroraccumulationedgeintelligenceCIFAR-10
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 federated learning over cellular networks can be made dramatically faster by adding a second level of hierarchy and by keeping every wireless message sparse. Mobile users in each small cell send only their largest gradient entries to a nearby small-cell base station; those base stations run several local rounds before exchanging model updates with a macro base station for global consensus. On CIFAR-10 with ResNet18, the scheme reports large per-iteration latency speedups over flat federated learning, with the speedup growing as the number of local rounds increases and as path loss becomes more severe, while the final Top-1 accuracy stays close to centralized training and slightly above flat FL. A careful reader would care because the dominant cost in real federated learning is often the wireless link, and this is a concrete protocol-level answer to that bottleneck that does not require sharing raw data.

What carries the argument

The load-bearing object is a hierarchical SGD protocol with two levels of parameter servers: each SBS aggregates and broadcasts gradients within its cluster for $H$ iterations, then the MBS averages the cluster models to restore global consensus. The protocol's per-iteration latency is derived from $\Gamma_{\mathrm{HFL}} = \Gamma_{\mathrm{period}}/H$, where $\Gamma_{\mathrm{period}} = \max_n ( \sum_{i=1}^{H} \Gamma_n^U(i) + \Gamma_n^D(i) ) + \Theta^U + \Theta^D + \max_n \Gamma_n^D$. Around that core sit three supporting mechanisms: the truncated channel-inversion power and rate model with max-min sub-carrier allocation (Algorithm 2, proved optimal by Theorem 1), gradient sparsification with momentum correction, and discounted error accumulation with discount factors $\beta_m$ and $\beta_s$.

What would settle it

Run the same CIFAR-10/ResNet18 training on a testbed or full simulator that includes per-iteration local computation time at the mobile users, and compare flat FL against HFL for the same sparsification parameters and values of $H$. The paper's central claim predicts HFL reaches a target accuracy in substantially less wall-clock time; if HFL is no faster, or slower, once computation is counted, the claimed end-to-end latency reduction would not hold in practice.

Watch

Extended reading notes

Core claim

The central claim, stated in the paper's own terms, is that a two-tier federation—small-cell base stations acting as intra-cluster parameter servers and a macro base station acting as the global aggregator every $H$ rounds, with sparsification on all four communication links and discounted error accumulation—reduces per-iteration communication latency in heterogeneous cellular networks by a large factor without degrading the learned model. The CIFAR-10/ResNet18 experiments report Top-1 accuracy of $90.27 \pm 0.11$ for $H=2$, $90.474 \pm 0.20$ for $H=4$, and $91.03 \pm 0.19$ for $H=6$, compared with $89.23 \pm 0.42$ for flat FL and $92.48 \pm 0.13$ for the centralized baseline. The same experiments report latency speedups that improve with $H$ and with the path-loss exponent, and sparsification multiplies the speedup for both schemes.

Load-bearing premise

The load-bearing premise is that the time to finish an iteration is just uplink plus downlink transmission time, with the time devices spend computing gradients taken to be negligible; if local computation time is significant, the reported latency reductions may not translate into faster wall-clock training.

Editorial extensions

If this is right

  • Longer local-update periods increase the latency speedup: the reported speedup grows from $H=2$ to $H=6$, and $H=6$ also gives the best reported accuracy among the hierarchical variants.
  • The speedup over flat federated learning grows with the path-loss exponent, so the hierarchical scheme is most valuable where signal attenuation is severe, such as dense urban or indoor deployments.
  • The max-min sub-carrier allocation makes per-iteration uplink latency equal to the slowest user's upload time, so the resource allocation scheme works by removing the bottleneck user rather than by improving average throughput.
  • Flat federated learning in the experiments also uses downlink sparsification and reaches lower accuracy than the hierarchical scheme, which the paper interprets as evidence that the hierarchy compensates for information lost in sparsified broadcasts.

Reading between the lines

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

  • The paper's latency model omits local computation time and the authors list an extended analysis with computation as future work; my inference is that in compute-bound regimes the hierarchy's advantage shrinks because iteration time becomes dominated by gradient calculation rather than transmission.
  • The reported accuracy difference between the hierarchical and flat schemes may depend on the flat baseline also using downlink sparsification; my inference is that a flat baseline without downlink sparsification would clarify how much of the accuracy gap comes from the hierarchy itself rather than from the sparsified broadcast.
  • The optimality proof for the greedy sub-carrier allocation is tied to the paper's specific rate model; my inference is that the same max-min greedy policy would remain a reasonable heuristic under other channel models, though its optimality would need to be re-established.
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

4 major / 5 minor

Summary. The paper proposes a hierarchical federated learning (HFL) framework for heterogeneous cellular networks, in which mobile users (MUs) send gradient updates to small-cell base stations (SBSs) for intra-cluster aggregation, and SBSs periodically communicate with a macro base station (MBS) for global model averaging. Gradient sparsification and periodic averaging are used to reduce communication volume, and a greedy subcarrier allocation algorithm is proposed to minimize uplink latency. The paper presents a communication-latency analysis for flat FL and HFL, and reports CIFAR-10/ResNet18 experiments showing that HFL achieves lower per-iteration latency than FL while maintaining or improving accuracy.

Significance. If the central claims hold, the paper makes a useful contribution to communication-efficient federated edge learning by jointly addressing hierarchical aggregation, sparsification, and wireless resource allocation in a realistic multi-tier topology. The latency analysis is transparent and the authors provide a public code repository, which supports reproducibility. However, because the quantitative latency speedup is computed from a model that omits local computation, and because the accuracy comparison is confounded by sparsification choices, the practical significance of the reported gains is not fully established.

major comments (4)
  1. [II-B, III-A, V-D] The per-iteration latency is defined as TFL = TUL + TDL (Section II-B) and ΓHFL = Γperiod/H (Section III-A), where Γperiod in Eq. (21) counts only transmission over the four links; no term accounts for local minibatch computation or server-side aggregation. Since the learning procedure is synchronous, wall-clock time per iteration is T_compute + T_comm, and the reported speedup TFL/ΓHFL assumes T_compute = 0. If computation is not negligible, the wall-clock speedup approaches 1, and the paper's own Section V-D lists training time with MU computation as future work. The authors should either restrict the claims to communication latency (and adjust the 'end-to-end' language in the contribution list) or augment the model and experiments with computation time.
  2. [II-C, Theorem 1] The proof of the optimality of the greedy subcarrier allocation in Theorem 1 is incomplete. The key inequality in the base case, \(\bar U_{k*}(2) > \bar U_{k*}(1) = \min_k \bar U^*_k\), asserts an equality for the alternative policy that is not demonstrated, and the induction step assumes that the optimal allocation of K+m subcarriers coincides with the greedy allocation without proof. The proof also ignores the threshold optimization in Eq. (11) that is part of the algorithm. Since the reported latency values rely on this allocation, the authors should supply a rigorous exchange argument (or state the greedy policy as a heuristic).
  3. [V-C, Table III, Figure 6] The accuracy comparison does not include a no-sparsification HFL baseline, and the FL baseline is run with downlink sparsification that the authors themselves identify as the likely cause of FL's poor accuracy. Consequently, the claim that HFL achieves better accuracy than FL may be an artifact of the asymmetry in sparsification rather than of the hierarchical architecture. The paper should add a fair baseline: HFL without sparsification and FL with the same sparsification levels on all links (or at least without downlink sparsification), and ideally report accuracy as a function of the sparsity factors.
  4. [V-B, V-C] The reported accuracy/latency tradeoff depends on hand-picked values of the sparsification parameters (φul_MU=0.99, φdl_SBS=0.9, φul_SBS=0.9, φdl_MBS=0.9) and discount factors (β_m=0.2, β_s=0.5), with no sensitivity analysis. Without such an analysis, it is difficult to assess how robust the 'without sacrificing accuracy much' conclusion is to these choices. A small sweep over these parameters, or a discussion of their effect, would strengthen the claims.
minor comments (5)
  1. [IV, Eq. (22)] In the text following Eq. (21), the error buffer is written as v_{k,t} = v_{k-1,t} + g_{k,t}, which appears to be a typo; it should be v_{k,t} = v_{k,t-1} + u_{k,t} as in Algorithm 4.
  2. [V-B] The stated baseline batch size K×β=128 is inconsistent with the reported values K=28 MUs and β=64, which give a global batch of 1792; please clarify.
  3. [Algorithm 5] Line 26 of Algorithm 5 reads 'en(t + 1) =' with the right-hand side on the following line; this formatting error should be fixed.
  4. [Table III] The caption 'Baseline result is obtained by training a single MU on the whole training set' should presumably read 'a single machine' or 'centralized training'.
  5. [V-C] The sentence 'This is due to the fact that MBS is serving more number of MUs than a single MBS' is confusing, since in HFL the MBS serves SBSs rather than MUs directly; the sentence should be rephrased.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: latency is computed from an explicit network model and accuracy is measured from independent CIFAR-10 runs.

full rationale

The paper's central claims are that hierarchical FL with sparsification reduces communication latency and preserves accuracy. The latency speedup is not a fitted or renamed input: it is the output of the communication-theoretic expressions in Sections II-A, II-B, and III-A (T_FL = T_UL + T_DL, Gamma_period = max_n(sum_i Gamma_U_n(i) + Gamma_D_n(i)) + Theta_U + Theta_D + max_n Gamma_D_n, speedup = T_FL / Gamma_HFL). The accuracy results in Figure 6 and Table III are obtained from actual ResNet-18 training runs on CIFAR-10, compared against a baseline and FL; the sparsification ratios and discount factors beta_m, beta_s are user-set hyperparameters, not parameters fitted to force either the latency or accuracy conclusions. The only notable gap is the acknowledged omission of MU computation time from the per-iteration latency model (Section II-B, with an extended training-time analysis deferred to Section V-D). That is a modeling limitation affecting the validity or generality of the wall-clock speedup claim, but it is not circularity: omitting a term does not make a derived quantity equivalent to its inputs by construction. Self-citations such as [1], [9], and [11] appear only as background references and are not load-bearing for the main derivation. No uniqueness theorem is invoked, and no result is defined in terms of the quantity it is supposed to predict. Accordingly, the paper shows no significant circularity.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

The central latency and accuracy claims depend on a stylized network model and several hand-chosen hyperparameters; the discount factors and sparsification levels are not derived from first principles, and the latency model assumes negligible computation and zero inter-cluster interference.

free parameters (4)
  • beta_m = 0.2
    Discount factor for error accumulation at the MBS (Algorithm 5); chosen by hand; affects accuracy and convergence; no sensitivity analysis reported.
  • beta_s = 0.5
    Discount factor for error accumulation at SBSs (Algorithm 5); chosen by hand; affects accuracy and convergence; no sensitivity analysis reported.
  • Sparsification parameters = phi_ul_MU=0.99; phi_dl_SBS=0.9; phi_ul_SBS=0.9; phi_dl_MBS=0.9
    Fixed across all experiments; control the fraction of gradients/models transmitted and directly trade off latency against accuracy; not derived or swept in the paper.
  • Learning rate schedule = initial 0.25, warm-up 5 epochs, drop by 10x at epochs 150 and 225
    Standard large-batch training recipe from [28], chosen by hand; determines reported accuracy but not central to latency claim.
assumptions (5)
  • domain assumption Assumption 1: MUs are uniformly distributed and each cluster contains equal number of MUs.
    Used in the HFL latency analysis (Section III-A) to make cluster-level aggregation symmetric; real HCN topologies are non-uniform.
  • domain assumption Assumption 2: SBSs are located at the origin of the corresponding clusters.
    Used to compute distances and path loss in the latency model (Section III-A).
  • domain assumption Zero interference from clusters beyond distance Dth through frequency reuse coloring.
    Latency analysis assumes perfect isolation between same-color clusters (Section III-A); real systems have residual interference.
  • domain assumption Truncated channel inversion with fixed-rate M-QAM from [17] describes the MU-SBS and MU-MBS links.
    Adopted from prior work; assumes known channel pdf and target BER, and does not model scheduling or retransmissions.
  • domain assumption Local computation time at MUs is negligible in the per-iteration latency.
    TFL is defined as TUL + TDL (Section II-B) and the paper defers computation time to future work (Section V-D).

how reviews work

0 comments
Cite this review

Pith. "Pith review of Hierarchical Federated Learning Across Heterogeneous Cellular Networks." pith.science (2026). https://pith.science/paper/B46BPFDR

@misc{pith2026190902362,
  author       = {Pith},
  title        = {Pith review of: Hierarchical Federated Learning Across Heterogeneous Cellular Networks},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/B46BPFDR}},
  note         = {Machine review of arXiv:1909.02362}
}
read the original abstract

We study collaborative machine learning (ML) across wireless devices, each with its own local dataset. Offloading these datasets to a cloud or an edge server to implement powerful ML solutions is often not feasible due to latency, bandwidth and privacy constraints. Instead, we consider federated edge learning (FEEL), where the devices share local updates on the model parameters rather than their datasets. We consider a heterogeneous cellular network (HCN), where small cell base stations (SBSs) orchestrate FL among the mobile users (MUs) within their cells, and periodically exchange model updates with the macro base station (MBS) for global consensus. We employ gradient sparsification and periodic averaging to increase the communication efficiency of this hierarchical federated learning (FL) framework. We then show using CIFAR-10 dataset that the proposed hierarchical learning solution can significantly reduce the communication latency without sacrificing the model accuracy.

Figures

Figures reproduced from arXiv: 1909.02362 by the authors.

Figure 1
Figure 1. Hierarchical FL cluster n is denoted by Γ U n = maxk QQˆ U¯ ∗n k , ∀k ∈ Cn. Similarly, let R¯∗n k the maximum average DL rate of MU k ∈ Cn. The DL latency of gradient aggregation in cluster n is denoted by Γ D n = maxk QQˆ R¯∗n k , ∀k ∈ Cn. After H iterations, SBSs send the model to the MBS for the purpose of averaging the clusters model. Let U SBS, RSBS be the UL, DL rate of SBSs to the MBS, respectively. The UL, D… view at source ↗
Figure 2
Figure 2. Clustering layout. Frequency reuse pattern is one. Each [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Latency Speed Up HFL versus FL. drop the initial learning rate by factor of 10, and similarly end of the 225th epoch we drop the learning rate by factor of 10 again. Further, for the weight decay factor3 and the momentum term we use w = 0.0001 and σ = 0.9 respectively in all simulations. Finally, for the discounted error accumulation we set βm = 0.2 and βs = 0.5. C. Results We first study the amount of speed up in l… view at source ↗
Figures from the paper (2 more)
Figure 6
Figure 6. Figure 6: Top-1 accuracy [6] K. Bonawitz, H. Eichner, W. Grieskamp, D. Huba, A. Ingerman, V. Ivanov, C. Kiddon, J. Konecny, S. Mazzocchi, H. B. McMahan, ´ T. V. Overveldt, D. Petrou, D. Ramage, and J. Roselander, “Towards federated learning at scale: System design,” CoRR, vol. a…
Figure 5
Figure 5. Figure 5: Latency speed up due to sparsification REFERENCES [1] M. M. Amiri and D. Gund ¨ uz, “Machine learning at the wireless ¨ edge: Distributed stochastic gradient descent over-the-air,” CoRR, vol. abs/1901.00844, 2019. [2] K. Yang, T. Jiang, Y. Shi, and Z. Ding, “Federated …

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

32 extracted references · 18 canonical work pages

  1. [1]

    Machine learning at the wireless edge: Distributed stochastic gradient descent over-the-air,

    M. M. Amiri and D. G ¨und¨uz, “Machine learning at the wireless edge: Distributed stochastic gradient descent over-the-air,” CoRR, vol. abs/1901.00844, 2019

  2. [2]

    Federated learning via over- the-air computation,

    K. Yang, T. Jiang, Y . Shi, and Z. Ding, “Federated learning via over- the-air computation,” CoRR, vol. abs/1812.11750, 2018

  3. [3]

    Low-latency broadband analog aggregation for federated edge learning,

    G. Zhu, Y . Wang, and K. Huang, “Low-latency broadband analog aggregation for federated edge learning,” CoRR, vol. abs/1812.11494, 2018

  4. [4]

    Wireless network intelligence at the edge,

    J. Park, S. Samarakoon, M. Bennis, and M. Debbah, “Wireless network intelligence at the edge,” CoRR, vol. abs/1812.02858, 2018. [Online]. Available: http://arxiv.org/abs/1812.02858

  5. [5]

    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 Proceedings of the 20th International Conference on Artificial Intelligence and Statistics , ser. Proceedings of Machine Learning Research, A. Singh and J. Zhu, Eds., vol. 54. Fort Lauderdale, FL, USA: PMLR, 20–22 A...

  6. [7]

    Communicating or computing over the mac: Function-centric wireless networks,

    L. Chen, N. Zhao, Y . Chen, F. R. Yu, and G. Wei, “Communicating or computing over the mac: Function-centric wireless networks,” IEEE Transactions on Communications , pp. 1–1, 2019

  7. [8]

    Towards an intelligent edge: Wireless communication meets machine learning,

    G. Zhu, D. Liu, Y . Du, C. You, J. Zhang, and K. Huang, “Towards an intelligent edge: Wireless communication meets machine learning,” CoRR, vol. abs/1809.00343, 2018. [Online]. Available: http://arxiv.org/abs/1809.00343

  8. [9]

    Machine learning in the air,

    D. G ¨und¨uz, P. de Kerret, N. D. Sidiropoulos, D. Gesbert, C. Murthy, and M. van der Schaar, “Machine learning in the air,” CoRR, vol. abs/1904.12385, 2019. [Online]. Available: http://arxiv.org/abs/1904. 12385

Show all 32 references
  1. [10]

    Wireless federated distillation for distributed edge learning with heterogeneous data,

    J.-H. Ahn, O. Simeone, and J. Kang, “Wireless federated distillation for distributed edge learning with heterogeneous data,” CoRR, vol. abs/1907.02745, 2019

  2. [11]

    Collaborative machine learning at the wireless edge with blind transmitters,

    M. M. Amiri, T. M. Duman, and D. G ¨und¨uz, “Collaborative machine learning at the wireless edge with blind transmitters,” CoRR, vol. abs/1907.03909, 2019. [Online]. Available: http://arxiv.org/abs/1907. 03909

  3. [12]

    Accelerating DNN training in wireless federated edge learning system,

    J. Ren, G. Yu, and G. Ding, “Accelerating DNN training in wireless federated edge learning system,” CoRR, vol. abs/1905.09712, 2019

  4. [13]

    Federated learning over wireless networks: Optimization model design and analysis,

    N. H. Tran, W. Bao, A. Zomaya, N. Minh N.H., and C. S. Hong, “Federated learning over wireless networks: Optimization model design and analysis,” inIEEE INFOCOM 2019 - IEEE Conference on Computer Communications, April 2019, pp. 1387–1395

  5. [14]

    Don’t use large mini-batches, use local SGD,

    T. Lin, S. U. Stich, and M. Jaggi, “Don’t use large mini-batches, use local SGD,” CoRR, vol. abs/1808.07217, 2018

  6. [15]

    A distributed hierarchical SGD algorithm with sparse global reduction,

    F. Zhou and G. Cong, “A distributed hierarchical SGD algorithm with sparse global reduction,” CoRR, vol. abs/1903.05133, 2019

  7. [16]

    Edge-assisted hierarchical federated learning with non-iid data,

    L. Liu, J. Zhang, S. H. Song, and K. B. Letaief, “Edge-assisted hierarchical federated learning with non-iid data,” CoRR, vol. abs/1905.06641, 2019. [Online]. Available: http://arxiv.org/abs/1905. 06641

  8. [17]

    Variable-rate variable-power mqam for fading channels,

    A. J. Goldsmith and S.-G. Chua, “Variable-rate variable-power mqam for fading channels,” IEEE Transactions on Communications , vol. 45, no. 10, pp. 1218–1230, Oct 1997

  9. [18]

    Adaptive federated learning in resource constrained edge com- puting systems,

    S. Wang, T. Tuor, T. Salonidis, K. K. Leung, C. Makaya, T. He, and K. Chan, “Adaptive federated learning in resource constrained edge com- puting systems,” IEEE Journal on Selected Areas in Communications , vol. 37, no. 6, pp. 1205–1221, June 2019

  10. [19]

    Deep gradient compression: Reducing the communication bandwidth for distributed training,

    Y . Lin, S. Han, H. Mao, Y . Wang, and B. Dally, “Deep gradient compression: Reducing the communication bandwidth for distributed training,” in International Conference on Learning Representations , 2018

  11. [20]

    Communication compression for decentralized training,

    H. Tang, S. Gan, C. Zhang, T. Zhang, and J. Liu, “Communication compression for decentralized training,” in Advances in Neural Infor- mation Processing Systems 31 , S. Bengio, H. Wallach, H. Larochelle, K. Grauman, N. Cesa-Bianchi, and R. Garnett, Eds. Curran Associates, Inc.,...

  12. [21]

    Robust and communication-efficient federated learning from non-iid data,

    F. Sattler, S. Wiedemann, K. M ¨uller, and W. Samek, “Robust and communication-efficient federated learning from non-iid data,” CoRR, vol. abs/1903.02891, 2019. [Online]. Available: http://arxiv.org/abs/ 1903.02891

  13. [22]

    Advanced mobile phone service: The cellular con- cept,

    V . H. M. Donald, “Advanced mobile phone service: The cellular con- cept,” The Bell System Technical Journal, vol. 58, no. 1, pp. 15–41, Jan 1979

  14. [23]

    Exploring 5g fronthaul network architecture intelligence splits and connectivity,

    T. Koelling and J. Rogers, “Exploring 5g fronthaul network architecture intelligence splits and connectivity,” March 2019, White Paper. [Online]. Available: https: //www.intel.co.uk/content/www/uk/en/architecture-and-technology/ silicon-photonics/5g-fronthaul-network-architect...

  15. [24]

    Infso-ict-247733 earth deliverable d 2 . 3 energy efficiency analysis of the reference systems , areas of improvements and target breakdown,

    W. Wajda, G. Auer, P. Skillermark, and Y . Jading, “Infso-ict-247733 earth deliverable d 2 . 3 energy efficiency analysis of the reference systems , areas of improvements and target breakdown,” 2012

  16. [25]

    Learning multiple layers of features from tiny images,

    A. Krizhevsky, “Learning multiple layers of features from tiny images,” Tech. Rep., 2009

  17. [26]

    Deep residual learning for image recognition,

    K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” in 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), June 2016, pp. 770–778

  18. [27]

    Hierarchical FL,

    M. S. H. Abad, “Hierarchical FL,” https://github.com/msalehihabad/ hierarchicalFL, 2019

  19. [28]

    Accurate, large minibatch SGD: training imagenet in 1 hour,

    P. Goyal, P. Doll ´ar, R. B. Girshick, P. Noordhuis, L. Wesolowski, A. Kyrola, A. Tulloch, Y . Jia, and K. He, “Accurate, large minibatch SGD: training imagenet in 1 hour,” CoRR, vol. abs/1706.02677, 2017. [Online]. Available: http://arxiv.org/abs/1706.02677

  20. [29]

    On the linear speedup analysis of communication efficient momentum SGD for distributed non-convex optimization,

    H. Yu, R. Jin, and S. Yang, “On the linear speedup analysis of communication efficient momentum SGD for distributed non-convex optimization,” in Proceedings of the 36th International Conference on Machine Learning , ser. Proceedings of Machine Learning Research, K. Chaudhuri an...

  21. [30]

    Federated learning with non-iid data,

    Y . Zhao, M. Li, L. Lai, N. Suda, D. Civin, and V . Chandra, “Federated learning with non-iid data,” CoRR, vol. abs/1806.00582, 2018. [Online]. Available: http://arxiv.org/abs/1806.00582

  22. [31]

    Communication-efficient on-device machine learning: Federated distillation and augmentation under non-iid private data,

    E. Jeong, S. Oh, H. Kim, J. Park, M. Bennis, and S. Kim, “Communication-efficient on-device machine learning: Federated distillation and augmentation under non-iid private data,” CoRR, vol. abs/1811.11479, 2018. [Online]. Available: http://arxiv.org/abs/1811.11479

  23. [32]

    Measuring the effects of data parallelism on neural network training,

    C. J. Shallue, J. Lee, J. M. Antognini, J. Sohl-Dickstein, R. Frostig, and G. E. Dahl, “Measuring the effects of data parallelism on neural network training,” CoRR, vol. abs/1811.03600, 2018. [Online]. Available: http://arxiv.org/abs/1811.03600

  24. [2019]

    Available: http://arxiv.org/abs/1902.01046

    [Online]. Available: http://arxiv.org/abs/1902.01046

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.