Pith. sign in

REVIEW 3 major objections 5 minor 1 cited by

NebulaFL: Effective Asynchronous Federated Learning for JointCloud Computing

T0 review · 3 major / 5 minor · reviewed 2026-08-11 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read NebulaFL claims that asynchronous version-controlled federated learning across multiple cloud data centers can improve accuracy, halve inter-center communication, and cut resource cost.

desk verdict Solid systems paper with a plausible empirical story, but the convergence proof is invalid and should be withdrawn or repaired; worth peer review with major revision expected. read the letter →

arxiv 2412.04868 v1 pith:DTRQQZ7F submitted 2024-12-06 cs.DC cs.AIcs.NI

Fei Gao , Ming Hu , Zhiyu Xie , Peichang Shi , Xiaofei Xie , Guodong Yi , Huaimin Wang This is my paper · ORCID
classification cs.DCcs.AIcs.NI
keywords asynchronousfederatedlearningJointCloudcomputingasaserviceversioncontrolmodelrotationresourceschedulingnon-IIDdataTEEcontainers
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

The paper is trying to establish that federated learning can be made practical across multiple cloud data centers, where heterogeneous data, cross-center communication, and resource pricing all work against standard FL. It proposes NebulaFL, which runs asynchronous training on several 'planet' models inside each data center, exchanges models between centers by rotating a 'stellar' model, and assigns containers and hardware by a reward-guided schedule. On the paper's own terms, the evidence is empirical—up to 5.71% accuracy improvement, up to 50% reduction in communication overhead, and up to 61.94% lower cost to reach a target accuracy—backed by a convergence bound modeled on FedAvg's analysis. If right, it would give cloud providers a concrete route to offering privacy-preserving collaborative learning across TEE-protected data owner containers without frequent all-to-all model aggregation.

What carries the argument

The three named objects are planet models (intermediate models trained asynchronously inside a data center), a stellar model (the last master model received from a peer center, used as a knowledge-sharing anchor), and a master model (the version-weighted aggregate of a center's planet models, sent to one peer each rotation round). The load-bearing updates are $\mathrm{PlanetUpdate}(m_j^i, m_s^i, v_j^i, v_s^i) = \frac{w_j^i m_j^i + d_t m_s^i}{w_j^i + d_t}$ with $w_j^i = \max(v_j^i - v_s^i, 5)$ and decaying $d_t \in [0.5, 1.0]$, plus the version-weighted master generation $\mathrm{MasterGen}(S_i, V_i) = \frac{\sum_k m_k^i v_k^i}{\sum_k v_k^i}$. The rotation schedule sends the master model to center $(i+r) \bmod |C| + 1$ on the $r$-th round, which is what cuts per-round communication. Container selection combines a performance reward with a curiosity reward from MBIE-EB, and resource assignment minimizes a cost factor plus a time factor.

What would settle it

Run the exact update sequence in Equations (10)-(12) on a one-dimensional quadratic loss with a fixed step size, a stale stellar model, and $\epsilon$ strictly between 0 and 1; if for any batch realization the empirical mean $E w_t$ exceeds the unblended average $v_t$, then Lemma 4's inequality $E w_t \le v_t$ fails and the proof of Theorem 1 breaks.

Watch

Extended reading notes

Core claim

On its own terms, NebulaFL's central discovery is that combining version control with a decentralized rotation of master models can replace the all-to-all aggregation step that dominates cross-cloud federated learning communication. Each data center keeps multiple planet models trained asynchronously; when a planet finishes a round, it is blended with a stellar model received from another center, and the center periodically sends its version-weighted master model to one designated peer in rotation. The paper argues that this preserves and often improves accuracy relative to FedAvg and four state-of-the-art baselines while transmitting only one master model per center per rotation round, which is the source of the claimed 50% communication saving. The reward-guided container selection and cost-time resource scheduling are presented as the components that let a low-version model get a fast container and a high-version model get a cheap container, balancing training time against rental cost.

Load-bearing premise

The formal convergence guarantee depends on a specific inequality stating that mixing with the stellar model shrinks the expected model parameters below the plain average; if that inequality does not hold, the paper provides no rigorous reason that the rotation mixture must converge.

Editorial extensions

If this is right

  • If the empirical results hold, FLaaS providers can let data owners train on TEE-backed cloud containers while sharing only one master model per center per rotation round, reducing cross-cloud traffic by about half per round compared with all-to-all aggregation.
  • Non-IID data distributed across centers does not have to force synchronous global rounds; NebulaFL reaches target accuracy with up to 61.94% lower rental cost and up to 63.41% faster training time in the reported ResNet-18/CIFAR-10 settings.
  • Longer rotation intervals slow early convergence while very short intervals risk premature convergence, making the rotation timer a tuneable trade-off between communication frequency and final accuracy.
  • Ablation results indicate that ignoring either the time factor or the cost factor in resource scheduling degrades training time or cost, so the paper claims both factors should be optimized together.
  • Increasing the number of data centers or the number of active containers tends to widen NebulaFL's accuracy advantage over the baselines, which the paper interprets as evidence of scalability.

Reading between the lines

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

  • The same version-control-plus-rotation pattern would transfer naturally to wide-area or satellite-linked FL where a central parameter server is unavailable, though the paper does not explore that setting.
  • The claimed 50% communication saving is relative to all-to-all aggregation each round; under sparse or partial inter-center topologies the actual saving would be smaller, a quantification the paper does not provide.
  • The convergence proof's Lemma 4 is the piece to check first: an independent verification or counterexample on the exact update equations would settle whether the stated bound applies, while the empirical comparisons would stand either way.
  • Because resource prices and runtimes are simulated from one platform's listed prices, a live deployment with current cloud GPU pricing would be a natural test of whether the 61.94% cost saving survives outside the simulator.
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 / 5 minor

Summary. The paper proposes NebulaFL, an asynchronous federated learning framework for multi-data-center JointCloud settings. It combines three mechanisms: version-controlled multiple planet models within each data center to mitigate data heterogeneity and stragglers, a decentralized model-rotation mechanism for inter-data-center knowledge sharing to reduce communication overhead, and a reward-guided container selection and resource scheduling strategy to balance training time and cost. The authors report experiments on CIFAR-10, CIFAR-100, and Fashion-MNIST with CNN, ResNet-18, and VGG-16, claiming up to 5.71% accuracy improvement, 50% communication reduction, and 61.94% cost savings under a target accuracy, and they include a convergence analysis in Section III-C.

Significance. If the empirical results are reliable, NebulaFL is a useful systems contribution to federated learning in multi-cloud, TEE-based settings. The evaluation is broad: five baselines, three models, three datasets, multiple non-IID levels, and standard deviations on accuracy, plus ablations for the scheduling and rotation designs; these are genuine strengths. The theoretical section, however, is not currently a valid derivation: the central Lemma 4 is unsupported and Theorem 1 does not follow from the stated updates. The contribution is therefore primarily empirical, and the theory must either be fixed or explicitly de-emphasized before the paper can be accepted.

major comments (3)
  1. [Section III-C, Lemma 4] The stated identity E[w_t] = (epsilon m)/(m+epsilon-1) v_t does not follow from the updates in Eqs. (10)-(12). With Eq. (11), w_k^{t+1} = epsilon v_k^t + (1-epsilon) S, so averaging over k yields w_{t+1} = epsilon v_t + (1-epsilon) S and hence E[w_{t+1}] = epsilon E[v_t] + (1-epsilon) E[S]. The lemma has no dependence on the stellar model S, and at epsilon=0 it predicts E[w_{t+1}]=0 while Eq. (11) gives w_{t+1}=S. Because Lemma 4 is the only bridge between the stellar-model mixing and the FedAvg-style bound, Theorem 1 and Eq. (13) are not established for NebulaFL.
  2. [Section III-C, Assumption 4 vs Algorithm 1] The convergence analysis assumes simple averaging of m randomly selected containers (Assumption 4), but Algorithm 1 uses version-weighted MasterGen (Eq. 3) and PlanetUpdate (Eq. 2), which mixes each planet model with the stellar model using version-dependent weights and a decaying weight d_t. Even if Lemma 4 were corrected, the proven bound would apply to a simplified average-based scheme rather than to the implemented algorithm. The authors should either extend the analysis to the version-weighted updates or state explicitly that the theoretical result covers only a simplified variant.
  3. [Section III-C, Theorem 1] The bound in Eq. (13) is the FedAvg bound from [33] with the same learning rate and constants, and the only NebulaFL-specific modification is the unproven Lemma 4. The sentence 'This is anticipated, as our method converges faster than FedAvg' is an assertion, not a consequence of the derivation. Since Lemma 4 is unsupported, the claimed smaller convergence bound has no basis; the authors should supply a correct argument or remove the convergence claim and present Section III-C as a heuristic discussion.
minor comments (5)
  1. [Section IV-A] The dataset name 'Fashion-MINIST' should be 'Fashion-MNIST'.
  2. [Section I] The phrase 'simi-aynchronous' is a typo; 'semi-asynchronous' is intended.
  3. [Section III-B, Eq. (9)] The formula for F_time appears to be missing a closing parenthesis; the intended grouping should be clarified.
  4. [Table III] The ablation results are reported without standard deviations or repeated runs; adding statistics would strengthen the scheduling ablation.
  5. [Figure 2] In the manuscript text provided, the figure content appears as garbled typesetting; please check the final PDF for label legibility.

Circularity Check

1 steps flagged · score 5.0 of 10

Theorem 1's 'smaller convergence bound' is not derived: Lemma 4 asserts Ew_t=(εm)/(m+ε−1)v_t≤v_t with only the justification that faster convergence is 'anticipated,' so the formal claim assumes its own conclusion; empirical claims remain externally benchmarked.

  1. self definitional [Section III-C (Convergence Analysis), Lemma 4 and Theorem 1, Eqs. (10)-(13)]
    "Specifically, Ewt = εm/(m+ε−1) vt ≤ vt, where ε ∈ [0, 1]. This is anticipated, as our method converges faster than FedAvg [18]. Consequently, wt is not an unbiased estimate of vt, allowing us to achieve a smaller convergence bound compared to [33]."

    The claimed prediction — 'a smaller convergence bound compared to [33]' — rests entirely on Lemma 4, whose identity is not a consequence of the paper's own updates: averaging Eq. (11) over k gives E[w_{t+1}] = εE[v_t] + (1−ε)E[S], which contradicts the lemma at ε=0 (Eq. (11) yields the stellar model, the lemma yields 0), and the lemma's factor never references the stellar model appearing in Eq. (11). The only justification offered is that faster convergence is 'anticipated'; the conclusion is the premise. Theorem 1 then restates the FedAvg bound of [33] with the asserted inequality supplying the claimed improvement, so the formal result reduces to an assertion of itself rather than a derivation from Eqs. (10)-(12).

full rationale

The paper's headline results — up to 5.71% accuracy gain, up to 50% communication reduction, and up to 61.94% cost savings — are empirical measurements against external baselines (FedAvg, FedHiSyn, FedHKT, FedSA, Pisces) with ablations against randomized or component-ablated variants, so the central experimental claims are self-contained and no data-fitted parameter is relabeled as a prediction. The heavy reuse of GitFL [14], a co-authored prior paper, for the intra-data-center machinery (multiple planet models, version-guided master aggregation, container selection, curiosity reward) is explicitly and repeatedly attributed ('Inspired by [14]', 'which is still used in [14]'), and GitFL is not invoked as the proof of any measured claim; this is transparency about inheritance rather than load-bearing self-citation. The convergence section legitimately imports the FedAvg convergence machinery of the external reference [33]; the mismatch between Assumption 4 (simple averaging) and the implemented PlanetUpdate/MasterGen is a correctness gap, not circularity. The one exhibited circular step is in Section III-C: Lemma 4's identity Ew_t=(εm)/(m+ε−1)v_t≤v_t does not follow from Eqs. (10)-(12) — averaging Eq. (11) gives an unavoidable stellar-model term, and at ε=0 the lemma predicts 0 while the equation gives the stellar model — and the only stated support is that faster convergence is 'anticipated,' i.e., the conclusion is the premise of Lemma 4, which then produces Theorem 1's smaller bound. Because this circularity affects only the formal convergence claim (one supporting section) while the empirical core is externally benchmarked, the score is 5 rather than 6-8.

Assumptions & free parameters 5 free parameters · 3 assumptions · 2 invented entities

The central claims of accuracy, cost, and communication efficiency rest on the simulated JointCloud environment and on the unsupported convergence lemma. The algorithm itself reuses GitFL's version-control scheme with a new rotation layer, and the cost results depend on free parameters that are not fully disclosed.

free parameters (5)
  • stellar decay weight d_t = range [0.5, 1.0]
    Hand-picked decaying weight in Eq. (2) controlling how much the stellar model influences planet models; decreases over time and resets at rotation. No sensitivity analysis is given beyond rotation-time ablation.
  • minimum planet weight floor = 5
    Lower bound on w_i^j in Eq. (2) to avoid excessive knowledge exchange; set by hand without justification.
  • rotation interval t_c = not stated for main experiments; ablated over 500, 1000, 1500, 2000
    Interval between inter-DC model rotations (Algorithm 1 input and Fig. 5b); the default value used in Tables I-II and Figures 2-4 is not reported.
  • runtime simulation distribution parameters = not specified
    Section IV-A: runtime of each resource is simulated based on a normal distribution with no mean or variance given; this directly determines the reported time and cost savings.
  • cost-time weighting in ResAssign = implicit equal weighting
    Eqs. (8)-(9) are summed with equal weight; no trade-off parameter is given, yet balancing time and cost is a stated contribution.
assumptions (3)
  • domain assumption Assumptions 1-4 from [33]: f_k is L-smooth, μ-strongly convex, stochastic gradient variance and norm are bounded, and containers are randomly sampled with replacement.
    Section III-C states these assumptions without justification for the neural networks used in experiments (CNN, ResNet-18, VGG-16), which are non-convex.
  • ad hoc to paper Lemma 4 (unbiasedness modification): Ew_t = (epsilon m)/(m+epsilon-1) v_t <= v_t.
    Section III-C: asserted to enable a smaller convergence bound than [33]; no proof is given, and the relation is not derived from Eqs. (10)-(12).
  • domain assumption The simulated normal runtime distribution accurately represents real resource heterogeneity.
    Section IV-A: all cost and time results depend on this simulated runtime; no real multi-cloud testbed is used.
invented entities (2)
  • Stellar model
    purpose: A per-data-center knowledge-sharing model updated by master models received from other data centers during rotation; used to pull planet models toward shared knowledge.
    Algorithmic construct introduced by the paper; its benefit is shown only through the paper's experiments and not independently verified outside the paper.
  • Planet model
    purpose: One of K intermediate models per data center; assigned to containers for asynchronous local training with version-based selection.
    Renamed version of GitFL's branch model concept; no external evidence specific to this naming or instantiation.

how reviews work

0 comments
Cite this review

Pith. "Pith review of NebulaFL: Effective Asynchronous Federated Learning for JointCloud Computing." pith.science (2026). https://pith.science/paper/DTRQQZ7F

@misc{pith2026241204868,
  author       = {Pith},
  title        = {Pith review of: NebulaFL: Effective Asynchronous Federated Learning for JointCloud Computing},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/DTRQQZ7F}},
  note         = {Machine review of arXiv:2412.04868}
}
read the original abstract

With advancements in AI infrastructure and Trusted Execution Environment (TEE) technology, Federated Learning as a Service (FLaaS) through JointCloud Computing (JCC) is promising to break through the resource constraints caused by heterogeneous edge devices in the traditional Federated Learning (FL) paradigm. Specifically, with the protection from TEE, data owners can achieve efficient model training with high-performance AI services in the cloud. By providing additional FL services, cloud service providers can achieve collaborative learning among data owners. However, FLaaS still faces three challenges, i.e., i) low training performance caused by heterogeneous data among data owners, ii) high communication overhead among different clouds (i.e., data centers), and iii) lack of efficient resource scheduling strategies to balance training time and cost. To address these challenges, this paper presents a novel asynchronous FL approach named NebulaFL for collaborative model training among multiple clouds. To address data heterogeneity issues, NebulaFL adopts a version control-based asynchronous FL training scheme in each data center to balance training time among data owners. To reduce communication overhead, NebulaFL adopts a decentralized model rotation mechanism to achieve effective knowledge sharing among data centers. To balance training time and cost, NebulaFL integrates a reward-guided strategy for data owners selection and resource scheduling. The experimental results demonstrate that, compared to the state-of-the-art FL methods, NebulaFL can achieve up to 5.71\% accuracy improvement. In addition, NebulaFL can reduce up to 50% communication overhead and 61.94% costs under a target accuracy.

Figures

Figures reproduced from arXiv: 2412.04868 by the authors.

Figure 1
Figure 1. Framework and workflow of our NebulaFL approach [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Comparison of communication overhead with different configs [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Learning curves for different numbers of centers [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Learning curves for different numbers of centers [PITH_FULL_IMAGE:figures/full_fig_p009_4.png]
Figure 5
Figure 5. Figure 5: Ablation study aboout rotation strategy 2) Rotation Strategy Ablation: Considering the communi￾cation overhead between data centers, we adopted a periodic rotation strategy to reduce the number of communications while ensuring knowledge sharing between different data c…

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Gradients as an Action: Towards Communication-Efficient Federated Recommender Systems via Adaptive Action Sharing

    cs.LG 2025-07 conditional novelty 6.0 of 10

    FedRAS clusters item-embedding gradients into a few shared centroids, called actions, and transmits those instead of full embeddings, cutting federated recommender communication costs by up to 96.88%.

Reference graph

Works this paper leans on

43 extracted references · 32 canonical work pages · cited by 1 Pith paper

  1. [33]

    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,” in International Conference on Learning Representations, 2020

    work page 2020

  2. [1]

    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, pp. 1273– 1282

    work page 2017

  3. [2]

    Federated optimization in heterogeneous networks,

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

    work page 2020

  4. [3]

    Fedmut: Generalized federated learning via stochastic mutation,

    M. Hu, Y . Cao, A. Li, Z. Li, C. Liu, T. Li, M. Chen, and Y . Liu, “Fedmut: Generalized federated learning via stochastic mutation,” in Proc. of the AAAI Conference on Artificial Intelligence , vol. 38, no. 11, 2024, pp. 12 528–12 537

    work page 2024

  5. [4]

    Quantitative timing analysis for cyber-physical systems using uncertainty-aware scenario- based specifications,

    M. Hu, W. Duan, M. Zhang, T. Wei, and M. Chen, “Quantitative timing analysis for cyber-physical systems using uncertainty-aware scenario- based specifications,” IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems , vol. 39, no. 11, pp. 4006–4017, 2020

    work page 2020

  6. [5]

    Trusted execution en- vironment: What it is, and what it is not,

    M. Sabt, M. Achemlal, and A. Bouabdallah, “Trusted execution en- vironment: What it is, and what it is not,” in Proc. of IEEE Trust- com/BigDataSE/Ispa, vol. 1, 2015, pp. 57–64

    work page 2015

  7. [6]

    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 Proc. of International conference on machine learning. PMLR, 2020, pp. 5132–5143

    work page 2020

  8. [7]

    Jointcloud: A cross-cloud cooperation architecture for integrated internet service customization,

    H. Wang, P. Shi, and Y . Zhang, “Jointcloud: A cross-cloud cooperation architecture for integrated internet service customization,” in 2017 IEEE 37th international conference on distributed computing systems (ICDCS). IEEE, 2017, pp. 1846–1855

    work page 2017

Show all 43 references
  1. [8]

    Cor- poration architecture for multiple cloud service providers in jointcloud computing,

    P. Shi, H. Wang, X. Yue, S. Yang, S. Yang, X. Fu, and Y . Peng, “Cor- poration architecture for multiple cloud service providers in jointcloud computing,” in 2017 IEEE 37th International Conference on Distributed Computing Systems Workshops (ICDCSW) . IEEE, 2017, pp. 294–298

    2017

  2. [9]

    Preservation of the global knowledge by not-true distillation in federated learning,

    G. Lee, M. Jeong, Y . Shin, S. Bae, and S.-Y . Yun, “Preservation of the global knowledge by not-true distillation in federated learning,” Ad- vances in Neural Information Processing Systems , vol. 35, pp. 38 461– 38 474, 2022

    2022

  3. [10]

    Fedcluster: A federated learning framework for cross-device private ecg classification,

    D. Lin, Y . Guo, H. Sun, and Y . Chen, “Fedcluster: A federated learning framework for cross-device private ecg classification,” in IEEE Confer- ence on Computer Communications Workshops (INFOCOM WKSHPS) . IEEE, 2022, pp. 1–6

    2022

  4. [11]

    Fedcross: Towards accurate federated learning via multi- model cross-aggregation,

    M. Hu, P. Zhou, Z. Yue, Z. Ling, Y . Huang, A. Li, Y . Liu, X. Lian, and M. Chen, “Fedcross: Towards accurate federated learning via multi- model cross-aggregation,” in IEEE International Conference on Data Engineering (ICDE). IEEE, 2024, pp. 2137–2150

    2024

  5. [12]

    Is aggregation the only choice? federated learning via layer- wise model recombination,

    M. Hu, Z. Yue, X. Xie, C. Chen, Y . Huang, X. Wei, X. Lian, Y . Liu, and M. Chen, “Is aggregation the only choice? federated learning via layer- wise model recombination,” in Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, 2024, pp. 1096– 1107

    2024

  6. [13]

    Asynchronous federated optimization,

    C. Xie, S. Koyejo, and I. Gupta, “Asynchronous federated optimization,” arXiv preprint arXiv:1903.03934 , 2019

    1903 arXiv

  7. [14]

    Gitfl: Uncertainty-aware real-time asynchronous federated learning using version control,

    M. Hu, Z. Xia, D. Yan, Z. Yue, J. Xia, Y . Huang, Y . Liu, and M. Chen, “Gitfl: Uncertainty-aware real-time asynchronous federated learning using version control,” in Proc. of IEEE Real-Time Systems Symposium (RTSS), 2023, pp. 145–157

    2023

  8. [15]

    Safa: A semi- asynchronous protocol for fast federated learning with low overhead,

    W. Wu, L. He, W. Lin, R. Mao, C. Maple, and S. Jarvis, “Safa: A semi- asynchronous protocol for fast federated learning with low overhead,” IEEE Transactions on Computers , vol. 70, no. 5, pp. 655–668, 2020

    2020

  9. [16]

    Fedsa: A semi-asynchronous federated learning mechanism in heterogeneous edge computing,

    Q. Ma, Y . Xu, H. Xu, Z. Jiang, L. Huang, and H. Huang, “Fedsa: A semi-asynchronous federated learning mechanism in heterogeneous edge computing,” IEEE Journal on Selected Areas in Communications , vol. 39, no. 12, pp. 3654–3672, 2021

    2021

  10. [17]

    Scaling distributed machine learning with the parameter server,

    M. Li, D. G. Andersen, J. W. Park, A. J. Smola, A. Ahmed, V . Josifovski, J. Long, E. J. Shekita, and B.-Y . Su, “Scaling distributed machine learning with the parameter server,” in 11th USENIX Symposium on operating systems design and implementation (OSDI 14), 2014, pp. 583– 598

    2014

  11. [18]

    Local sgd with periodic averaging: Tighter analysis and adaptive synchro- nization,

    F. Haddadpour, M. M. Kamani, M. Mahdavi, and V . Cadambe, “Local sgd with periodic averaging: Tighter analysis and adaptive synchro- nization,” Advances in Neural Information Processing Systems , vol. 32, 2019

    2019

  12. [19]

    Fedrelay: Federated relay learning for 6g mobile edge intelligence,

    P. Li, Y . Zhong, C. Zhang, Y . Wu, and R. Yu, “Fedrelay: Federated relay learning for 6g mobile edge intelligence,” IEEE Transactions on Vehicular Technology, vol. 72, no. 4, pp. 5125–5138, 2022

    2022

  13. [20]

    Privacy for free: Wireless federated learning via uncoded transmission with adaptive power control,

    D. Liu and O. Simeone, “Privacy for free: Wireless federated learning via uncoded transmission with adaptive power control,” IEEE Journal on Selected Areas in Communications, vol. 39, no. 1, pp. 170–185, 2020

    2020

  14. [21]

    Distributed learning in hetero- geneous environment: federated learning with adaptive aggregation and computation reduction,

    J. Li, T. Mahmoodi, and H.-K. Lam, “Distributed learning in hetero- geneous environment: federated learning with adaptive aggregation and computation reduction,” in ICC 2023-IEEE International Conference on Communications. IEEE, 2023, pp. 1976–1981

    2023

  15. [22]

    Adaptive configuration for heterogeneous participants in decentralized federated learning,

    Y . Liao, Y . Xu, H. Xu, L. Wang, and C. Qian, “Adaptive configuration for heterogeneous participants in decentralized federated learning,” in IEEE INFOCOM 2023-IEEE Conference on Computer Communications. IEEE, 2023, pp. 1–10

    2023

  16. [23]

    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

    2023

  17. [24]

    Helios: Heterogeneity-aware fed- erated learning with dynamically balanced collaboration,

    Z. Xu, F. Yu, J. Xiong, and X. Chen, “Helios: Heterogeneity-aware fed- erated learning with dynamically balanced collaboration,” in ACM/IEEE Design Automation Conference (DAC) . IEEE, 2021, pp. 997–1002

    2021

  18. [25]

    Model pruning enables efficient federated learning on edge devices,

    Y . Jiang, S. Wang, V . Valls, B. J. Ko, W.-H. Lee, K. K. Leung, and L. Tassiulas, “Model pruning enables efficient federated learning on edge devices,” IEEE Transactions on Neural Networks and Learning Systems, vol. 34, no. 12, pp. 10 374–10 386, 2022

    2022

  19. [26]

    Synchronize only the immature parameters: Communication-efficient federated learning by freezing parameters adaptively,

    C. Chen, H. Xu, W. Wang, B. Li, B. Li, L. Chen, and G. Zhang, “Synchronize only the immature parameters: Communication-efficient federated learning by freezing parameters adaptively,” IEEE Transac- tions on Parallel and Distributed Systems , 2023

    2023

  20. [27]

    User- distribution-aware federated learning for efficient communication and fast inference,

    Y . Cui, Z. Zhang, N. Wang, L. Li, C. Chang, and T. Wei, “User- distribution-aware federated learning for efficient communication and fast inference,” IEEE Transactions on Computers , 2023

    2023

  21. [28]

    Communication-efficient federated learning for heterogeneous edge devices based on adaptive gradient quantization,

    H. Liu, F. He, and G. Cao, “Communication-efficient federated learning for heterogeneous edge devices based on adaptive gradient quantization,” in IEEE INFOCOM 2023-IEEE Conference on Computer Communica- tions. IEEE, 2023, pp. 1–10

    2023

  22. [29]

    Enabling communication-efficient federated learning via distributed compressed sensing,

    Y . Guan, X. Liu, T. Ren, and J. Niu, “Enabling communication-efficient federated learning via distributed compressed sensing,” in IEEE INFO- COM 2023-IEEE Conference on Computer Communications . IEEE, 2023, pp. 1–10

    2023

  23. [30]

    Curiosity-driven exploration by self-supervised prediction,

    D. Pathak, P. Agrawal, A. A. Efros, and T. Darrell, “Curiosity-driven exploration by self-supervised prediction,” in International conference on machine learning . PMLR, 2017, pp. 2778–2787

    2017

  24. [31]

    Accelerating reinforcement learning-based ccsl specification synthesis using curiosity- driven exploration,

    M. Hu, M. Zhang, F. Mallet, X. Fu, and M. Chen, “Accelerating reinforcement learning-based ccsl specification synthesis using curiosity- driven exploration,” IEEE Transactions on Computers , vol. 72, no. 5, pp. 1431–1446, 2022

    2022

  25. [32]

    Unifying count-based exploration and intrinsic motivation,

    M. Bellemare, S. Srinivasan, G. Ostrovski, T. Schaul, D. Saxton, and R. Munos, “Unifying count-based exploration and intrinsic motivation,” Advances in neural information processing systems , vol. 29, 2016

    2016

  26. [34]

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

    T. Lin, S. U. Stich, K. K. Patel, and M. Jaggi, “Don’t use large mini- batches, use local sgd,” arXiv preprint arXiv:1808.07217 , 2018

    2018 arXiv

  27. [35]

    Parallel restarted sgd with faster con- vergence and less communication: Demystifying why model averaging works for deep learning,

    H. Yu, S. Yang, and S. Zhu, “Parallel restarted sgd with faster con- vergence and less communication: Demystifying why model averaging works for deep learning,” in Proceedings of the AAAI conference on artificial intelligence, vol. 33, no. 01, 2019, pp. 5693–5700

    2019

  28. [36]

    Communication-efficient algorithms for statistical optimization,

    Y . Zhang, M. J. Wainwright, and J. C. Duchi, “Communication-efficient algorithms for statistical optimization,” Advances in neural information processing systems, vol. 25, 2012

    2012

  29. [37]

    Autodl market list,

    AutoDL, “Autodl market list,” https://www.autodl.com/market/list, ac- cessed: 2024-07-31

    2024

  30. [38]

    Learning multiple layers of features from tiny images,

    A. Krizhevsky, G. Hinton et al. , “Learning multiple layers of features from tiny images,” 2009

    2009

  31. [39]

    Fashion-mnist: a novel image dataset for benchmarking machine learning algorithms,

    H. Xiao, K. Rasul, and R. V ollgraf, “Fashion-mnist: a novel image dataset for benchmarking machine learning algorithms,” arXiv preprint arXiv:1708.07747, 2017

    2017 arXiv

  32. [40]

    Measuring the effects of non- identical data distribution for federated visual classification,

    T.-M. H. Hsu, H. Qi, and M. Brown, “Measuring the effects of non- identical data distribution for federated visual classification,” arXiv preprint arXiv:1909.06335, 2019

    1909 arXiv

  33. [41]

    Fedhisyn: A hierarchical synchronous federated learning framework for resource and data heterogeneity,

    G. Li, Y . Hu, M. Zhang, J. Liu, Q. Yin, Y . Peng, and D. Dou, “Fedhisyn: A hierarchical synchronous federated learning framework for resource and data heterogeneity,” in Proceedings of the 51st International Con- ference on Parallel Processing, 2022, pp. 1–11

    2022

  34. [42]

    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 Proc. of IEEE Conference on Computer Communications , 2023, pp. 1–10

    2023

  35. [43]

    Pisces: Efficient federated learning via guided asynchronous training,

    Z. Jiang, W. Wang, B. Li, and B. Li, “Pisces: Efficient federated learning via guided asynchronous training,” in Proc. of Symposium on Cloud Computing, 2022, pp. 370–385

    2022

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.