Pith. sign in

REVIEW 4 major objections 6 minor 41 references

Efficient Data Labeling and Optimal Device Scheduling in HWNs Using Clustered Federated Semi-Supervised Learning

T0 review · 4 major / 6 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read Clustered federated learning can label unlabeled device data through cluster-specific pseudo-labeling, saving up to 51% energy.

desk verdict A genuine CFL+SSL integration for HWNs with a load-bearing gap: the utility that drives model selection is never defined — worth a serious review round, not ready as written. read the letter →

arxiv 2412.17081 v1 pith:YYANUPCE submitted 2024-12-22 cs.DC

Moqbel Hamood , Abdullatif Albaseer , Mohamed Abdallah , Ala Al-Fuqaha This is my paper · ORCID
classification cs.DC
keywords clusteredfederatedlearningsemi-supervisedhierarchicalwirelessnetworkspseudo-labelingnon-IIDdataworkerselectionenergyefficiencyensemblemodels
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 argues that clustered federated learning, normally built on the assumption that every device has labeled data, can be extended to realistic hierarchical wireless networks where most data is unlabeled. The proposed framework, CFSL, uses the specialized models produced by clustering to pseudo-label unlabeled data, and it decides which model should label, when labeling should start, and which workers should be scheduled. The paper reports that this beats labeled-only CFL, CFL with random worker selection, and hierarchical federated semi-supervised learning on FEMNIST and CIFAR-10 under non-IID data, while cutting energy consumption by up to 51%. The contribution is a practical bridge between clustered federated learning and the unlabeled-data conditions of real deployments.

What carries the argument

The central machinery is the specialized model produced by clustered federated learning, together with the utility function $U(\theta_m, \mathcal{D}^u_i)$ that is assumed to rank how well each specialized model can label a worker's unlabeled data. The best-performing scheme selects the model that maximizes this utility; the ensemble scheme sets weights $\alpha_{i,m}$ from the same utility. A confidence threshold $\Phi$ gates which pseudo-labels are accepted into the training set, and the split-based and stopping-based timing schemes decide when this labeling machinery turns on. The paper assumes this utility is computable without ground-truth labels, which is the premise that makes the best-performing scheme meaningful.

What would settle it

Reproduce the FEMNIST best-performing-specialized-model experiment while forcing every worker to use only its own cluster's model instead of a utility-ranked one; if labeling accuracy stays the same, model selection is not the source of the reported gains, and if accuracy collapses, the utility ranking is doing the work.

Watch

Extended reading notes

Core claim

CFSL's central claim is that the specialized models formed by clustering devices with similar data distributions can be repurposed as labelers for unlabeled, unseen data in a hierarchical wireless network. Two prediction-model schemes are proposed: assigning each worker the best-performing specialized model, and forming a weighted-averaging ensemble of all specialized models. Two timing schemes decide when labeling begins — split-based, as soon as any cluster split occurs, and stopping-based, only after clusters stabilize — and two scheduling schemes, greedy and round-robin, choose which workers continue training. The joint labeling-scheduling problem is formulated as an intractable mixed-integer nonlinear program and replaced by heuristic subproblems, and experiments report large gains in testing and labeling accuracy over labeled-only CFL, CFL with random worker selection, and hierarchical FL with SSL.

Load-bearing premise

The schemes assume workers can score how well each specialized model would label their unlabeled data without ever seeing true labels, but the paper never defines how that score is computed.

Editorial extensions

If this is right

  • If the results hold, clustered federated learning can operate when only 5–15% of device data is labeled, removing the main practical objection to CFL.
  • The stopping-based timing scheme implies that delaying pseudo-labeling until clusters stabilize improves labeling accuracy, while the split-based scheme trades accuracy for lower resource use.
  • The reported energy savings imply that intelligent worker scheduling plus early use of pseudo-labels can roughly halve training energy in hierarchical wireless networks.
  • The convergence analysis treats pseudo-label noise as a bias term controlled by a regularization weight and a variance term that shrinks with cluster size, so larger clusters should tolerate less reliable labelers.

Reading between the lines

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

  • The utility function is never given an explicit formula, so implementing the best-performing scheme requires inventing a label-free scoring rule; prediction confidence or consistency across augmentations are natural candidates worth testing.
  • The experiments use simulated channels and fixed path-loss models; an over-the-air deployment with real stragglers and fading would show whether the 51% energy saving persists.
  • The paper does not combine the two model schemes, so a hybrid — best model for generating pseudo-labels, ensemble for final prediction — is an untested extension that the components suggest.
  • The convergence proof assumes strong convexity and smoothness, which convolutional networks and VGG-19 do not satisfy, so the stated convergence rate is a heuristic bound for the actual experiments.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 6 minor

Summary. The paper proposes CFSL, a framework that combines clustered federated learning (CFL) with semi-supervised pseudo-labeling in hierarchical wireless networks (HWNs). It introduces two prediction-model schemes (best-performing specialized model and weighted-averaging ensemble), two prediction-timing schemes (split-based and stopping-based), and two worker-selection strategies (greedy and round-robin), which are combined into eight integrated scheme variants. The authors formulate a joint optimization problem P1, reformulate it as P2/P3 for the two prediction schemes, provide a convergence analysis in Section VI, and report FEMNIST and CIFAR-10 experiments claiming large improvements in testing and labeling accuracy and up to 51% energy savings over labeled-only CFL, CFL with SSL and random selection, and HFL with SSL baselines.

Significance. If the empirical claims are reproducible, the paper addresses a real gap: most CFL work assumes labeled data, whereas practical edge data are mostly unlabeled. The eight scheme combinations give a useful engineering taxonomy, and the computation/communication models reflect genuine system-level concerns. The paper also makes a good-faith attempt to formulate the joint labeling, timing, and scheduling problem. However, the central algorithm depends on an undefined utility function, the convergence analysis is not a proof under the actual algorithm, and the experiments omit essential implementation details. The framework is promising, but as written the headline gains cannot be traced to a fully specified method. The paper is not yet ready for publication in its current form.

major comments (4)
  1. [Section V-A, Eqs. (23a) and (24a)] The utility function U(θ_m, D^u_i) is never defined. It appears in both reformulated objectives P2 and P3, it determines which specialized model is selected as 'best-performing' in Section V-A.1, and it sets the ensemble weights α_{i,m} in Section V-A.2. Without a concrete formula or algorithm for computing U from unlabeled data, neither prediction scheme is implementable and the Section VII experiments cannot be reproduced. Please specify U explicitly (e.g., average max-softmax confidence, negative entropy, or another label-free statistic) and explain how each worker obtains and evaluates all M specialized models.
  2. [Section VI, Eq. (31)] The regularization term Reg_i(θ) is defined using the true labels y_z^{(i)} of unlabeled data. In the problem setting of this paper, unlabeled data have no ground-truth labels, so this term cannot be computed by any worker or server. The convergence analysis therefore analyzes a different objective from the one in P1–P3. Either redefine Reg_i using only pseudo-labels and a confidence mask, or explicitly state that the analysis applies to an idealized oracle that knows the true labels.
  3. [Section VI, Eqs. (40)–(45)] The convergence analysis is a sketch rather than a proof. After assuming L-smoothness and strong convexity, the paper writes a standard contraction inequality without proving it for the actual CFSL algorithm, which involves clustered splits, hierarchical aggregation, partial participation, and pseudo-label injection. The variance bound Var(Reg_i(θ)) ≤ σ_R^2 / I_m in Eq. (44) is asserted without derivation, and Eq. (45) does not follow from the preceding display because the interaction between the bias term (41) and the variance term (42) is not analyzed. Please either provide a theorem with explicit assumptions and a complete proof, or clearly label this as an informal heuristic argument and remove the claimed convergence-rate statement.
  4. [Section VII, Results and Discussion] The experiments do not report enough detail to support the quantitative headline claims. There are no standard deviations or seed information, no values for ε_1, ε_2, λ, ν, or the splitting thresholds, no description of how the model-selection step was instantiated in the simulations, and no statement of how many runs produced Figs. 6–11. In particular, the 123.33% mean-accuracy improvement in Fig. 6 and the 51% energy saving in Fig. 10 cannot be validated without the exact protocol. Please provide a reproducibility appendix (including hyperparameters and, ideally, code) or temper the claims accordingly.
minor comments (6)
  1. [Table I] The row for unlabeled data repeats the symbol D_l^i; it should be D_u^i to match the text in Section III-A.
  2. [Section III-C, Assumption 1] The set expression `SMJ n=1 gjn` is malformed; it should be the union ∪_{n=1}^{M_j} g_{jn}.
  3. [Algorithm 1, lines 7–9] The quantity γ_i and the symbol `simmaxcross` are not defined, and the splitting condition using them does not match conditions (5)–(6) in the text; the pseudocode should be aligned with the formal conditions.
  4. [Section VII-A and Table II] The labeled-data fraction is written as '0.5, 0.10, and 0.15' in the text and '0.5' in Table II; the intended values are clearly 0.05, 0.10, and 0.15, and the typo should be corrected.
  5. [Section V-D and Section VII-B5] The 'Lessons Learned' section claims the approach reduces time and energy costs, but the evaluation reports only energy consumption; either add training-time results or remove the time claim.
  6. [Algorithm 3, line 8] The condition `∥∇θ Fi(θ∗j )∥< ε2 > 0` is syntactically ambiguous and should be rewritten.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the paper's performance claims are empirical and its design choices are not reductions of their own inputs.

full rationale

I walked the paper's claimed derivation chain: problem formulation P1 (Eq. 22a), the heuristic reformulations P2 (Eq. 23a) and P3 (Eq. 24a), the prediction model and time schemes, the worker selection strategies, and the convergence analysis in Section VI. None of these steps equates a predicted result to a fitted input by construction. The pseudo-labeling procedure is a standard SSL loop controlled by a confidence threshold (Eqs. 17-18), not a circular definition. The utility function U(theta_m, D_u_i) that appears in P2 and P3 is never defined, and Eq. (31) defines the regularization term using true labels y_z for unlabeled data; these are under-specification and correctness/consistency gaps, but they are not cases where an output is equivalent to its input by definition. The paper's headline claims (e.g., 123.33% mean accuracy improvement and 51% energy savings) are empirical measurements against baselines, not predictions derived from fitted parameters. Self-citations such as [1], [23], [26], and [27] are used as related-work context and preliminary results; none is invoked as a load-bearing theorem or as a uniqueness argument that forces the proposed design. No imported uniqueness theorem from the authors appears. Hence, under the stated rubric, there is no significant circularity, and the appropriate score is 0.

Assumptions & free parameters 6 free parameters · 5 assumptions · 0 invented entities

The central performance claim rests on unverified idealizations: stationary-point gradient geometry, a defined-but-unavailable utility, and a convergence analysis that assumes true labels for unlabeled data. The paper also does not report values for several governing hyperparameters (epsilon_1, epsilon_2, nu, lambda) or release code, so the ledger is heavy on uncharged assumptions.

free parameters (6)
  • epsilon_1 = not reported
    Controls when the HFL objective is close to a stationary point in Eq. (5); no value or tuning procedure is given.
  • epsilon_2 = not reported
    Controls worker grouping in Eq. (6) and the stopping point in Eq. (25); no value is reported.
  • Phi (confidence threshold) = 0.6, 0.7, 0.8, 0.9 (tested sweep)
    Determines which pseudo-labels are accepted in Eq. (17); the paper reports results per value but gives no rule for choosing it.
  • nu (regularization coefficient) = not reported
    Controls the weight of the labeling-correction term in the convergence analysis, Eqs. (30), (33), and (39); no value or selection method is given.
  • lambda (trade-off parameter) = not reported
    Appears in the P2 and P3 objectives (Eqs. (23a) and (24a)) to balance loss and labeling utility; never specified.
  • ensemble weights alpha_i,m = not specified
    Adjusted heuristically based on an undefined utility function U in Section V-A.2; no update rule is provided.
assumptions (5)
  • domain assumption Cosine similarity between worker gradients at the stationary point is either 1 or -1 (Eq. 4).
    Inherited from CFL [15]; this idealizes stationary-point gradient geometry and is not verified for the CNN and VGG-19 models used in the experiments.
  • standard math Local loss functions are L-smooth, the global objective is mu-strongly convex, and stochastic gradient variance is bounded (Eqs. 34-36).
    Standard optimization assumptions invoked for the convergence analysis, but they are not justified for deep neural network losses and were not checked empirically.
  • ad hoc to paper The regularization term Reg_i is computable from true labels of unlabeled data (Eq. 31).
    The convergence analysis defines Reg_i as the gap between loss on predicted labels and loss on true labels, but true labels y_z for unlabeled data are unavailable by the problem setup, making the derivation inapplicable.
  • ad hoc to paper A utility function U(theta_m, D_u) exists and can be evaluated by workers on unlabeled data to rank specialized models or set ensemble weights.
    Central to the P2 and P3 objectives and to the best-performing model selection scheme, but the function is never defined or shown to be computable without labels.
  • domain assumption Workers in each edge network are generated from K distinct data distributions and can be partitioned into clusters satisfying CFL conditions (Assumption 1 and Eqs. 5-6).
    Imposes a clean mixture-of-distributions structure on non-IID data that may not hold in general deployments.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Efficient Data Labeling and Optimal Device Scheduling in HWNs Using Clustered Federated Semi-Supervised Learning." pith.science (2026). https://pith.science/paper/YYANUPCE

@misc{pith2026241217081,
  author       = {Pith},
  title        = {Pith review of: Efficient Data Labeling and Optimal Device Scheduling in HWNs Using Clustered Federated Semi-Supervised Learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/YYANUPCE}},
  note         = {Machine review of arXiv:2412.17081}
}
read the original abstract

Clustered Federated Multi-task Learning (CFL) has emerged as a promising technique to address statistical challenges, particularly with non-independent and identically distributed (non-IID) data across users. However, existing CFL studies entirely rely on the impractical assumption that devices possess access to accurate ground-truth labels. This assumption becomes problematic in hierarchical wireless networks (HWNs), with vast unlabeled data and dual-level model aggregation, slowing convergence speeds, extending processing times, and increasing resource consumption. To this end, we propose Clustered Federated Semi-Supervised Learning (CFSL), a novel framework tailored for realistic scenarios in HWNs. We leverage specialized models from device clustering and present two prediction model schemes: the best-performing specialized model and the weighted-averaging ensemble model. The former assigns the most suitable specialized model to label unlabeled data, while the latter unifies specialized models to capture broader data distributions. CFSL introduces two novel prediction time schemes, split-based and stopping-based, for accurate labeling timing, and two device selection strategies, greedy and round-robin. Extensive testing validates CFSL's superiority in labeling/testing accuracy and resource efficiency, achieving up to 51% energy savings.

Figures

Figures reproduced from arXiv: 2412.17081 by the authors.

Figure 1
Figure 1. The system model [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. System framework diagram for the CFSL approach. [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Illustrating labeling process for the CFSL in HWNs. [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figures from the paper (7 more)
Figure 4
Figure 4. Figure 4: Testing accuracy for the proposed approach compared [PITH_FULL_IMAGE:figures/full_fig_p013_4.png]
Figure 5
Figure 5. Figure 5: Testing accuracy for the proposed approach compared [PITH_FULL_IMAGE:figures/full_fig_p013_5.png]
Figure 6
Figure 6. Figure 6: Minimum, average, and maximum accuracies for different scenarios using FEMNIST. [PITH_FULL_IMAGE:figures/full_fig_p014_6.png]
Figure 7
Figure 7. Figure 7: Testing accuracy for the BMSPGS (CIFAR-10). [PITH_FULL_IMAGE:figures/full_fig_p014_7.png]
Figure 8
Figure 8. Figure 8: Testing accuracy for the EMSTGS (CIFAR-10). [PITH_FULL_IMAGE:figures/full_fig_p014_8.png]
Figure 10
Figure 10. Figure 10: Averaged energy consumption with split-based and stopping-based prediction time schemes at [PITH_FULL_IMAGE:figures/full_fig_p015_10.png]
Figure 11
Figure 11. Figure 11: Averaged energy consumption of BMSPGS (Greedy) vs. CFL with SSL (Random) at Φ = 0.70. and BMSPRR scenarios show energy savings of 55.98% and 53%, respectively. This significant saving arises from the efficient labeling of unlabeled data, accurate labeling time, and th…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

41 extracted references · 31 canonical work pages

  1. [1]

    Empow- ering hwns with efficient data labeling: A clustered federated semi- supervised learning approach,

    M. Hamood, A. Albaseer, M. Abdallah, and A. Al-Fuqaha, “Empow- ering hwns with efficient data labeling: A clustered federated semi- supervised learning approach,” in IEEE Wireless Communications and Networking Conference (WCNC) . IEEE, 2024. 16

    work page 2024

  2. [2]

    Sensor- based activity recognition,

    L. Chen, J. Hoey, C. D. Nugent, D. J. Cook, and Z. Yu, “Sensor- based activity recognition,” IEEE Transactions on Systems, Man, and Cybernetics, Part C (Applications and Reviews), vol. 42, no. 6, pp. 790– 808, 2012

    work page 2012

  3. [3]

    Human activity recognition with smartphone and wearable sensors using deep learning techniques: A review,

    E. Ramanujam, T. Perumal, and S. Padmavathi, “Human activity recognition with smartphone and wearable sensors using deep learning techniques: A review,” IEEE Sensors Journal , vol. 21, no. 12, pp. 13 029–13 040, 2021

    work page 2021

  4. [4]

    Deep learning for computer vision: A brief review,

    A. V oulodimos, N. Doulamis, A. Doulamis, E. Protopapadakis et al. , “Deep learning for computer vision: A brief review,” Computational intelligence and neuroscience , vol. 2018, 2018

    2018

  5. [5]

    A survey on vision transformer,

    K. Han, Y . Wang, H. Chen, X. Chen, J. Guo, Z. Liu, Y . Tang, A. Xiao, C. Xu, Y . Xuet al., “A survey on vision transformer,” IEEE transactions on pattern analysis and machine intelligence, vol. 45, no. 1, pp. 87–110, 2022

    2022

  6. [6]

    Application of machine learning in wireless networks: Key techniques and open issues,

    Y . Sun, M. Peng, Y . Zhou, Y . Huang, and S. Mao, “Application of machine learning in wireless networks: Key techniques and open issues,” IEEE Communications Surveys & Tutorials , vol. 21, no. 4, pp. 3072– 3108, 2019

    work page 2019

  7. [7]

    From federated to fog learning: Distributed machine learning over heterogeneous wireless networks,

    S. Hosseinalipour, C. G. Brinton, V . Aggarwal, H. Dai, and M. Chiang, “From federated to fog learning: Distributed machine learning over heterogeneous wireless networks,” IEEE Communications Magazine , vol. 58, no. 12, pp. 41–47, 2020

    work page 2020

  8. [8]

    Federated learning: Strategies for improving communication efficiency,

    J. Kone ˇcn`y, H. B. McMahan, F. X. Yu, P. Richt ´arik, A. T. Suresh, and D. Bacon, “Federated learning: Strategies for improving communication efficiency,” arXiv preprint arXiv:1610.05492 , 2016

    arXiv 2016

Show all 41 references
  1. [9]

    Hierarchical federated learning across heterogeneous cellular networks,

    M. S. H. Abad, E. Ozfatura, D. Gunduz, and O. Ercetin, “Hierarchical federated learning across heterogeneous cellular networks,” in ICASSP 2020-2020 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) . IEEE, 2020, pp. 8866–8870

    2020

  2. [10]

    Federated learning over wireless iot networks with optimized communication and resources,

    H. Chen, S. Huang, D. Zhang, M. Xiao, M. Skoglund, and H. V . Poor, “Federated learning over wireless iot networks with optimized communication and resources,” IEEE Internet of Things Journal , 2022

    2022

  3. [11]

    Fine-grained data selection for improved energy efficiency of federated edge learning,

    A. Albaseer, M. Abdallah, A. Al-Fuqaha, and A. Erbad, “Fine-grained data selection for improved energy efficiency of federated edge learning,” IEEE Transactions on Network Science and Engineering , vol. 9, no. 5, pp. 3258–3271, 2021

    2021

  4. [12]

    Hfel: Joint edge asso- ciation and resource allocation for cost-efficient hierarchical federated edge learning,

    S. Luo, X. Chen, Q. Wu, Z. Zhou, and S. Yu, “Hfel: Joint edge asso- ciation and resource allocation for cost-efficient hierarchical federated edge learning,” IEEE Transactions on Wireless Communications, vol. 19, no. 10, pp. 6535–6548, 2020

    2020

  5. [13]

    Federated learning for healthcare informatics,

    J. Xu, B. S. Glicksberg, C. Su, P. Walker, J. Bian, and F. Wang, “Federated learning for healthcare informatics,” Journal of Healthcare Informatics Research, vol. 5, pp. 1–19, 2021

    2021

  6. [14]

    Federated learning with soft cluster- ing,

    C. Li, G. Li, and P. K. Varshney, “Federated learning with soft cluster- ing,” IEEE Internet of Things Journal , vol. 9, no. 10, pp. 7773–7782, 2021

    2021

  7. [15]

    Clustered federated learning: Model-agnostic distributed multitask optimization under privacy con- straints,

    F. Sattler, K.-R. M ¨uller, and W. Samek, “Clustered federated learning: Model-agnostic distributed multitask optimization under privacy con- straints,” IEEE Transactions on Neural Networks and Learning Systems , vol. 32, no. 8, pp. 3710–3722, 2020

    2020

  8. [16]

    Three approaches for personalization with applications to federated learning,

    Y . Mansour, M. Mohri, J. Ro, and A. T. Suresh, “Three approaches for personalization with applications to federated learning,” arXiv preprint arXiv:2002.10619, 2020

    2002 arXiv

  9. [17]

    An efficient frame- work for clustered federated learning,

    A. Ghosh, J. Chung, D. Yin, and K. Ramchandran, “An efficient frame- work for clustered federated learning,” Advances in Neural Information Processing Systems, vol. 33, pp. 19 586–19 597, 2020

    2020

  10. [18]

    Client selection approach in support of clustered federated learning over wireless edge networks,

    A. Albaseer, M. Abdallah, A. Al-Fuqaha, and A. Erbad, “Client selection approach in support of clustered federated learning over wireless edge networks,” in 2021 IEEE Global Communications Conference (GLOBE- COM). IEEE, 2021, pp. 1–6

    2021

  11. [19]

    Energy-efficient clustering to address data heterogeneity in federated learning,

    Y . Luo, X. Liu, and J. Xiu, “Energy-efficient clustering to address data heterogeneity in federated learning,” in ICC 2021-IEEE International Conference on Communications . IEEE, 2021, pp. 1–6

    2021

  12. [20]

    On the convergence of clustered federated learning,

    J. Ma, G. Long, T. Zhou, J. Jiang, and C. Zhang, “On the convergence of clustered federated learning,” arXiv preprint arXiv:2202.06187, 2022

    2022 arXiv

  13. [21]

    Dynamic clustering in federated learning,

    Y . Kim, E. Al Hakim, J. Haraldson, H. Eriksson, J. M. B. da Silva, and C. Fischione, “Dynamic clustering in federated learning,” in ICC 2021- IEEE International Conference on Communications . IEEE, 2021, pp. 1–6

    2021

  14. [22]

    Adaptive client clustering for efficient federated learning over non-iid and imbalanced data,

    B. Gong, T. Xing, Z. Liu, W. Xi, and X. Chen, “Adaptive client clustering for efficient federated learning over non-iid and imbalanced data,” IEEE Transactions on Big Data , 2022

    2022

  15. [23]

    Intelligent model aggregation in hierarchical clustered federated mul- titask learning,

    M. Hamood, A. Albaseer, M. Abdallah, A. Al-Fuqaha, and A. Mohamed, “Intelligent model aggregation in hierarchical clustered federated mul- titask learning,” in GLOBECOM 2023-2023 IEEE Global Communica- tions Conference. IEEE, 2023, pp. 3009–3014

    2023

  16. [24]

    Active client selection for clustered federated learning,

    H. Huang, W. Shi, Y . Feng, C. Niu, G. Cheng, J. Huang, and Z. Liu, “Active client selection for clustered federated learning,” IEEE Trans- actions on Neural Networks and Learning Systems , 2023

    2023

  17. [25]

    Fair selection of edge nodes to participate in clustered federated multitask learning,

    A. Albaseer, M. Abdallah, A. Al-Fuqaha, A. Mohammed, A. Erbad, and O. A. Dobre, “Fair selection of edge nodes to participate in clustered federated multitask learning,” IEEE Transactions on Network and Service Management , 2023

    2023

  18. [26]

    Clustered and multi-tasked federated distillation for heterogeneous and resource constrained industrial iot applications,

    M. Hamood, A. Albaseer, M. Abdallah, and A. Al-Fuqaha, “Clustered and multi-tasked federated distillation for heterogeneous and resource constrained industrial iot applications,” IEEE Internet of Things Maga- zine, vol. 6, no. 2, pp. 64–69, 2023

    2023

  19. [27]

    Semi-supervised federated learning over heterogeneous wireless iot edge networks: Framework and algorithms,

    A. Albaseer, M. Abdallah, A. Al-Fuqaha, A. Erbad, and O. A. Dobre, “Semi-supervised federated learning over heterogeneous wireless iot edge networks: Framework and algorithms,” IEEE Internet of Things Journal, vol. 9, no. 24, pp. 25 626–25 642, 2022

    2022

  20. [28]

    Semifl: Semi-supervised federated learning for unlabeled clients with alternate training,

    E. Diao, J. Ding, and V . Tarokh, “Semifl: Semi-supervised federated learning for unlabeled clients with alternate training,” Advances in Neural Information Processing Systems , vol. 35, pp. 17 871–17 884, 2022

    2022

  21. [29]

    Semi-supervised and person- alized federated activity recognition based on active learning and label propagation,

    R. Presotto, G. Civitarese, and C. Bettini, “Semi-supervised and person- alized federated activity recognition based on active learning and label propagation,” Personal and Ubiquitous Computing , vol. 26, no. 5, pp. 1281–1298, 2022

    2022

  22. [30]

    Towards fast personalized semi-supervised federated learning in edge networks: Algorithm design and theoretical guarantee,

    S. Wang, Y . Xu, Y . Yuan, and T. Q. S. Quek, “Towards fast personalized semi-supervised federated learning in edge networks: Algorithm design and theoretical guarantee,” IEEE Transactions on Wireless Communica- tions, pp. 1–1, 2023

    2023

  23. [31]

    Semipfl: person- alized semi-supervised federated learning framework for edge intelli- gence,

    A. Tashakori, W. Zhang, Z. J. Wang, and P. Servati, “Semipfl: person- alized semi-supervised federated learning framework for edge intelli- gence,” IEEE Internet of Things Journal , 2023

    2023

  24. [32]

    Adaptive hier- archical federated learning over wireless networks,

    B. Xu, W. Xia, W. Wen, P. Liu, H. Zhao, and H. Zhu, “Adaptive hier- archical federated learning over wireless networks,” IEEE Transactions on Vehicular Technology, vol. 71, no. 2, pp. 2070–2083, 2021

    2021

  25. [33]

    Hierarchical federated learning with quantization: Convergence analysis and system design,

    L. Liu, J. Zhang, S. Song, and K. B. Letaief, “Hierarchical federated learning with quantization: Convergence analysis and system design,” IEEE Transactions on Wireless Communications , 2022

    2022

  26. [34]

    Auction-based cluster federated learning in mobile edge computing systems,

    R. Lu, W. Zhang, Y . Wang, Q. Li, X. Zhong, H. Yang, and D. Wang, “Auction-based cluster federated learning in mobile edge computing systems,” IEEE Transactions on Parallel and Distributed Systems , vol. 34, no. 4, pp. 1145–1158, 2023

    2023

  27. [35]

    Flexible clustered federated learning for client-level data distribution shift,

    M. Duan, D. Liu, X. Ji, Y . Wu, L. Liang, X. Chen, Y . Tan, and A. Ren, “Flexible clustered federated learning for client-level data distribution shift,” IEEE Transactions on Parallel and Distributed Systems , vol. 33, no. 11, pp. 2661–2674, 2021

    2021

  28. [36]

    Edge devices clustering for federated visual classification: A feature norm based framework,

    X.-X. Wei and H. Huang, “Edge devices clustering for federated visual classification: A feature norm based framework,” IEEE Transactions on Image Processing, vol. 32, pp. 995–1010, 2023

    2023

  29. [37]

    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 Artificial intelligence and statistics . PMLR, 2017, pp. 1273– 1282

    2017

  30. [38]

    Meta-gating framework for fast and continuous resource optimization in dynamic wireless environments,

    Q. Hou, M. Lee, G. Yu, and Y . Cai, “Meta-gating framework for fast and continuous resource optimization in dynamic wireless environments,” IEEE Transactions on Communications , vol. 71, no. 9, pp. 5259–5273, 2023

    2023

  31. [39]

    Convergence analysis of a distributed optimization algorithm with a general unbalanced directed communica- tion network,

    H. Li, Q. L ¨u, and T. Huang, “Convergence analysis of a distributed optimization algorithm with a general unbalanced directed communica- tion network,” IEEE Transactions on Network Science and Engineering , vol. 6, no. 3, pp. 237–248, 2019

    2019

  32. [40]

    Improving robustness using generated data,

    S. Gowal, S.-A. Rebuffi, O. Wiles, F. Stimberg, D. A. Calian, and T. A. Mann, “Improving robustness using generated data,” Advances in Neural Information Processing Systems , vol. 34, pp. 4218–4233, 2021

    2021

  33. [41]

    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. 17 Moqbel Hamood received his B.Sc. degree in Elec- trical Engineering from Mutah University,...

    2018 arXiv

Pith tools

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