Pith. sign in

REVIEW 3 major objections 6 minor 23 references

REDUS: Adaptive Resampling for Efficient Deep Learning in Centralized and Federated IoT Networks

T0 review · 3 major / 6 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read The paper claims that an AdaBoost-inspired per-epoch resampling scheme can cut deep-learning training time by 72.6% on the CICIoT2023 IoT attack dataset while reducing accuracy by only 1.62 percentage points.

desk verdict A plausible but unproven efficiency claim: the time model omits the full-dataset forward pass, so the 72.6% speedup is not supported by the algorithm as written. read the letter →

arxiv 2507.02021 v1 pith:DQEM6HZY submitted 2025-07-02 cs.NI

Eyad Gad , Gad Gad , Mostafa M. Fouda , Mohamed I. Ibrahem , Muhammad Ismail , Zubair Md Fadlullah This is my paper
classification cs.NI
keywords resamplingfederatedlearningdeepsoftware-definednetworkingIoTattackdetectionCICIoT2023datasettrainingtimereductionAdaBoost
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 proposes REDUS, an AdaBoost-inspired resampling method for deep learning training that each epoch keeps only samples whose running weight exceeds a threshold, so the model trains on a shrinking, hard-example-focused subset. On the CICIoT2023 IoT attack dataset, REDUS with threshold $\theta = 1.5\times 10^{-6}$ cuts average training time per round from 777 seconds to 213 seconds (72.60% reduction) while accuracy drops only 1.62 percentage points, from 94.33% to 92.71%. The authors argue that such targeted resampling balances computational demands and model performance, making deep learning feasible on resource-limited edge devices in centralized and federated settings. If true, REDUS offers a practical way to reduce training budgets in SDN-managed IoT networks without retraining models from scratch.

What carries the argument

The central object is the sample-weight update and thresholded inclusion mechanism in Algorithm 1. Each sample starts at weight $w_i = 1/n$; at every epoch after the first, the weighted misclassification rate $\varepsilon_t$ is computed, and every sample's weight is updated by an AdaBoost-style factor, increasing for misclassified samples and decreasing for correctly classified ones, then normalized by $Z_t$. Samples whose updated weight falls below $\theta$ are excluded from the next epoch's training set $\hat{D}$, but their weights keep being updated so they can re-enter if they become hard again. This mechanism converts an AdaBoost ensemble-building procedure into a per-epoch sample selector for a single deep network, defining the claimed trade-off between training time and accuracy.

What would settle it

Measure the wall-clock time of one REDUS epoch at $\theta = 1.5\times10^{-6}$ on CICIoT2023 with the same hardware and split it into the subset back-propagation time and the time spent computing $\varepsilon_t$ and updating all $n$ weights (including the full forward pass over excluded samples). If the latter is non-negligible, the observed 72.6% reduction relative to vanilla training will shrink correspondingly; a direct comparison of per-epoch times with and without the full-dataset forward pass would settle whether the claimed savings are attainable.

Watch

Extended reading notes

Core claim

The central claim is that a single strong deep-learning model can be trained efficiently by repeatedly reweighting samples with an AdaBoost-style update and discarding, each epoch, the samples whose weights fall below a fixed threshold $\theta$; the model is then trained only on the surviving subset $\hat{D}$. The weight of a misclassified sample is multiplied by $e^{\alpha_t}/Z_t$ and that of a correctly classified sample by $e^{-\alpha_t}/Z_t$, with $\alpha_t = \frac{1}{2}\log\frac{1-\varepsilon_t}{\varepsilon_t}$ and $Z_t = 2\sqrt{\varepsilon_t(1-\varepsilon_t)}$, so hard samples dominate the next epoch while already-learned samples drop out. Since the gradient updates are applied only to samples in $\hat{D}$, the per-epoch back-propagation cost falls as training proceeds, which the authors express through the time model $L_{TT} = |D| \cdot \tau \cdot E$ and the reported wall-clock savings. Evaluated on CICIoT2023 within a five-client federated setup, REDUS at the optimized thresholds achieves 59.84% and 72.60% training-time reductions with 0.68% and 1.62% accuracy reductions respectively, which the paper presents as evidence that the method preserves model performance while substantially cutting computational load.

Load-bearing premise

The method's claimed savings depend on the assumption that recomputing weights for all $n$ samples each epoch, including the excluded ones, is so cheap that total time is dominated by back-propagation on the surviving subset; if that full-dataset pass is counted, worst-case per-epoch cost remains linear in $n$ and the 72.6% savings do not follow as written.

Editorial extensions

If this is right

  • At $\theta = 1.3\times 10^{-6}$, training time drops 59.84% with only 0.68% accuracy loss, showing a favorable operating point.
  • At $\theta = 1.5\times 10^{-6}$, training time drops 72.60% with only 1.62% accuracy loss.
  • Because REDUS runs locally on each client before FedAvg aggregation, it reduces the per-client training time $L_{TT_i}$ and thus the round time in federated learning without changing communication rounds.
  • The threshold $\theta$ provides a tunable knob between computational savings and accuracy, with high thresholds ($1.7\times10^{-6}$ and above) cutting time by about 78% but losing 2.82 to 3.48% accuracy.
  • REDUS preserves the model's final accuracy within a couple of points on the tested dataset, supporting its use in resource-limited IoT and edge settings.

Reading between the lines

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

  • The reported time savings assume the cost of computing $\varepsilon_t$ and updating weights for all $n$ samples each epoch is negligible; if the full-dataset forward pass required for those updates is included, the worst-case per-epoch cost remains linear in $n$, and the savings depend on forward passes being much cheaper than back-propagation.
  • REDUS is essentially a hard-example mining schedule; one could test it against simpler heuristics like keeping a fixed top-$k$ fraction per epoch to see whether the AdaBoost weighting, rather than the subset selection, drives the observed gains.
  • The accuracy recovery at $\theta = 1.1\times10^{-6}$ (94.33%, same as baseline, with 0.52% time reduction) suggests non-monotonic behavior worth investigating; the re-inclusion dynamics may occasionally improve generalization.
  • In a federated setting, the threshold could be adapted per client based on local data difficulty or resource budget, but the paper does not explore this.
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

3 major / 6 minor

Summary. The paper proposes REDUS, an AdaBoost-inspired resampling technique that maintains per-sample weights during deep learning training and, at each epoch, trains only on the subset of samples whose weights exceed a threshold θ. The authors claim that this reduces the number of training samples per epoch, thereby reducing training time and energy consumption in both centralized and federated settings. The method is evaluated on the CICIoT2023 dataset for IoT attack detection in a federated learning setup, reporting a training time reduction of up to 72.6% at θ=1.5e-6 with an accuracy reduction of 1.62% relative to a baseline ANN. The paper also sketches an extension of REDUS to federated learning with FedAvg-style aggregation.

Significance. If the efficiency claim were rigorously supported, REDUS would be a simple and practical tool for reducing DL training budgets on resource-constrained IoT/edge devices, and the federated extension would be a natural fit for the paper's target scenario. The algorithmic idea is clearly stated in Algorithm 1, the equations are straightforward, and the use of a standard public dataset (CICIoT2023) is a strength. However, the manuscript does not provide code, timing logs, or a validation-based threshold selection procedure, and the central time-reduction argument omits an operation that the algorithm itself requires. As it stands, the headline result is not established, and the experimental protocol is circular in its selection of the reported operating point.

major comments (3)
  1. [Section IV-A, Eq. (6), and Algorithm 1 lines 8-13] The total-training-time model omits the full-dataset forward pass required by the algorithm. After the first epoch, computing ε_t in Eq. (9) and updating every w_i in Eqs. (10)-(11) requires evaluating f(x_i) for all i ∈ D, including samples excluded from the subset D_hat. Since Eq. (6) counts only |D|·τ·E, with τ defined as the back-propagation time per sample, the derived reduction is not justified. Including the forward pass, the per-epoch cost is n·t_f + |D_hat|·(t_f + t_b) for REDUS versus n·(t_f + t_b) for vanilla training; the maximum reduction when |D_hat|=0 is t_b/(t_f + t_b), which is only about 50-67% for typical t_b/t_f ratios of 2-3 and is lower for positive |D_hat|. The paper itself states two paragraphs below Eq. (11) that "the time complexity of the REDUS method is O(E · n)", which concedes the same asymptotic order as vanilla training. The measured times in Table I therefore need to be supported by an implementation description or timing profiles that account for the forward pass, or the algorithm and claims need to be revised.
  2. [Section V, Table I] The thresholds θ = 1.3×10^-6 and θ = 1.5×10^-6 are selected as "optimized" after evaluating test-set accuracy and time over the ten scanned threshold values. Because the same test evaluation is used both to select θ and to report the resulting accuracy, the advertised 1.62% accuracy reduction is a post-hoc selected value rather than an unbiased estimate of the method's generalization performance. The authors should use a validation split for threshold selection and then report test performance at the selected threshold, along with the variance across the five experimental repetitions.
  3. [Section IV-A, Eq. (10)] The weight-update rule is undefined when ε_t = 0 or ε_t = 1 because α_t = 0.5·log((1-ε_t)/ε_t) involves a division by zero and a logarithm of zero. Since REDUS progressively excludes correctly classified samples and concentrates on harder samples, a state with ε_t = 0 can plausibly occur after several epochs; the algorithm needs a fallback (for example, skip the weight update or set α_t = 0) to remain well-defined over the full training run.
minor comments (6)
  1. [Section III, Eq. (1)] The denominator is typeset as "P i = 1nwi" and should be Σ_{i=1}^n w_i; also, "wether" should be "whether".
  2. [Section IV-A, Eq. (6)] The symbols "LLT" and "LT T" appear to be typographical errors for the total training time L_TT.
  3. [Section IV-B, Eq. (12)] The summation index is written as K=1 to K; it should be i=1 to K.
  4. [Section V] The text says the threshold ranges from 0 to 1/n, but the experimental sweep goes from 0 to 2/3 × 1/n; please clarify the allowed range and the number of sampled steps.
  5. [Section V, Table I] No standard deviations or confidence intervals are reported for the five repetitions; because the accuracy differences of interest are small (e.g., 92.71% vs. 93.65%), variance information is needed to assess whether the reported accuracy drops are significant.
  6. [After Section VI (Fig. 4)] The text following the conclusion contains undecodable character sequences; the figure and its caption need to be regenerated in the manuscript.

Circularity Check

2 steps flagged · score 6.0 of 10

REDUS's headline speedup is an artifact of a backprop-only time definition that omits the full forward pass in Algorithm 1, plus a threshold selected from the same test-set evaluation.

  1. self definitional [Section IV-A, Eq. (6) and Algorithm 1, lines 8-15]
    "Then, the total training time (LLT ) is given by: LT T= |D| ·τ · E, (6) ... The goal of our proposed resampling approach is to gradually reduce the number of samples used in training during each epoch, thereby reducing the total training time as defined by LT T. ... The time complexity of the REDUS method is O(E · n)."

    Eq. (6) defines total training time solely as back-propagation on all |D| samples. Algorithm 1, however, evaluates f(x_i) for every sample in D each epoch (lines 8-13) to update weights, which requires a full forward pass not counted in L_TT. The claimed speedup is therefore exactly the reduction in the defined quantity, not a consequence of the algorithm's actual per-epoch cost; the paper's own O(E·n) complexity statement acknowledges the full pass remains. The 72.6% reduction is thus an artifact of defining time to exclude the one operation that the algorithm still performs on all samples.

  2. fitted input called prediction [Section V, Table I and following paragraph]
    "Threshold values for REDUS were sampled across 10 steps from 0 to 2/3 × 1/n ... Upon analysis, it's shown that the optimized threshold values θ = 1.3 × 10−6 and θ = 1.5 × 10−6 significantly reduce the training time by 59.84% and 72.60%, respectively ... The optimized thresholds ... are therefore ideal."

    The threshold θ is a free parameter scanned over ten values, and the two thresholds highlighted as 'optimized' are chosen from the same test-set evaluation whose accuracy and training time are then reported as REDUS's performance. The advertised trade-off (72.60% time reduction, 1.62% accuracy loss) is a post-hoc selected operating point rather than a parameter-free prediction. The conclusion that REDUS 'effectively balances computational demands and model performance' is inferred from the same data used to pick θ, so the headline result is a fitted input presented as an outcome of the method.

full rationale

The paper's central efficiency claim reduces to its own cost definition and to a selected operating point. Eq. (6) defines total training time as backprop-only on the full dataset, while Algorithm 1 still requires a full forward pass over D each epoch to update sample weights; the claimed savings follow from that definition, not from the algorithm's actual operations. Separately, the threshold θ is swept over ten values and the two best rows of Table I are then labeled 'optimized,' so the headline speedup/accuracy trade-off is chosen from the same test-set evaluation rather than predicted independently. The experimental benchmark is external (CICIoT2023) and the algorithm has independent content beyond self-citation; references [22] and [23] are self-citations but only supply preprocessing partitioning and future-work context, not the load-bearing derivation. Nonetheless, the main quantitative claim is partially circular: it depends on a definitional time model and on thresholds fitted to the reported evaluation.

Assumptions & free parameters 1 free parameters · 3 assumptions · 0 invented entities

The central claim rests on one tuned threshold θ and on two unproven modeling assumptions: that AdaBoost weights are meaningful for a single DNN, and that the full forward pass for weight updates is free. No new physical or conceptual entities are introduced. The threshold values 1.3e-6 and 1.5e-6 are selected from the test-set sweep, so the headline numbers are fitted operating points rather than parameter-free predictions.

free parameters (1)
  • threshold θ = 1.3e-6 and 1.5e-6 chosen as optimized; grid from 0 to (2/3) * (1/n)
    θ controls how many samples are dropped each epoch. It is swept over ten values on the CICIoT2023 test data and the best trade-off points are selected post hoc in Section V, Table I.
assumptions (3)
  • domain assumption Exponential AdaBoost-style weight updates remain valid for a single deep network trained with SGD, so low-weight samples are safely learned and can be excluded.
    Invoked in Algorithm 1, lines 8-13; no proof or ablation shows that the reweighted subset preserves the loss landscape or convergence behavior of the original objective.
  • ad hoc to paper Correctly classified samples below θ are redundant, and excluding them does not materially change accuracy.
    Core premise of Section IV-A; the paper only demonstrates this empirically for two selected thresholds and gives no theoretical support.
  • ad hoc to paper The full-dataset forward pass needed for weight updates is negligible relative to training on the selected subset.
    Needed for Eq. (6) to support the time-reduction claim; not stated, measured, or justified anywhere in the paper.

how reviews work

0 comments
Cite this review

Pith. "Pith review of REDUS: Adaptive Resampling for Efficient Deep Learning in Centralized and Federated IoT Networks." pith.science (2026). https://pith.science/paper/DQEM6HZY

@misc{pith2026250702021,
  author       = {Pith},
  title        = {Pith review of: REDUS: Adaptive Resampling for Efficient Deep Learning in Centralized and Federated IoT Networks},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/DQEM6HZY}},
  note         = {Machine review of arXiv:2507.02021}
}
read the original abstract

With the rise of Software-Defined Networking (SDN) for managing traffic and ensuring seamless operations across interconnected devices, challenges arise when SDN controllers share infrastructure with deep learning (DL) workloads. Resource contention between DL training and SDN operations, especially in latency-sensitive IoT environments, can degrade SDN's responsiveness and compromise network performance. Federated Learning (FL) helps address some of these concerns by decentralizing DL training to edge devices, thus reducing data transmission costs and enhancing privacy. Yet, the computational demands of DL training can still interfere with SDN's performance, especially under the continuous data streams characteristic of IoT systems. To mitigate this issue, we propose REDUS (Resampling for Efficient Data Utilization in Smart-Networks), a resampling technique that optimizes DL training by prioritizing misclassified samples and excluding redundant data, inspired by AdaBoost. REDUS reduces the number of training samples per epoch, thereby conserving computational resources, reducing energy consumption, and accelerating convergence without significantly impacting accuracy. Applied within an FL setup, REDUS enhances the efficiency of model training on resource-limited edge devices while maintaining network performance. In this paper, REDUS is evaluated on the CICIoT2023 dataset for IoT attack detection, showing a training time reduction of up to 72.6% with a minimal accuracy loss of only 1.62%, offering a scalable and practical solution for intelligent networks.

Figures

Figures reproduced from arXiv: 2507.02021 by the authors.

Figure 1
Figure 1. Overview of the Adaboost algorithm. a decision stump) is trained on the weighted training data. The error ϵt of an t iteration of the weak learner is calculated as: ϵt = Pn i=1 .wi .I(yi ̸= ht(xi)) Pi = 1nwi , (1) where I is a function that returns 1, if the sample xi is misclassified and 0 otherwise; ht(xi) is the prediction of the weak learner. • Step 3: The weights of the instances are updated based on the perfor… view at source ↗
Figure 2
Figure 2. An overview of the REDUS training. The figure shows training a [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Overview of the FL architecture adopted in this study. [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: The performance metrics achieved on the CICIoT2023 dataset with different threshold values. [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

23 extracted references · 22 canonical work pages

  1. [1]

    On softwarization of intelligence in 6G networks for ultra-fast optimal policy selection: Challenges and opportunities,

    S. Hashima, Z. M. Fadlullah, M. M. Fouda, E. M. Mohamed, K. Hatano, B. M. ElHalawany, and M. Guizani, “On softwarization of intelligence in 6G networks for ultra-fast optimal policy selection: Challenges and opportunities,” IEEE Network , vol. 37, no. 2, pp. 190–197, 2023

    work page 2023

  2. [2]

    Privacy-preserving data-driven learning mod- els for emerging communication networks: A comprehensive sur- vey,

    M. M. Fouda et al. , “Privacy-preserving data-driven learning mod- els for emerging communication networks: A comprehensive sur- vey,” IEEE Communications Surveys & Tutorials , early access, doi: 10.1109/COMST.2024.3486690

    arXiv 2024

  3. [3]

    A survey on federated learning,

    C. Zhang, Y . Xie, H. Bai, B. Yu, W. Li, and Y . Gao, “A survey on federated learning,” Knowledge-Based Systems , vol. 216, article no. 106775, 2021

    work page 2021

  4. [4]

    HCP: heterogeneous computing platform for federated learning based collaborative content caching towards 6g networks,

    Z. Md. Fadlullah and N. Kato, “HCP: heterogeneous computing platform for federated learning based collaborative content caching towards 6g networks,” IEEE Transactions on Emerging Topics in Computing, vol. 10, no. 1, pp. 112–123, 2022

    work page 2022

  5. [5]

    Privacy-preserving federated- learning-based net-energy forecasting,

    M. M. Badr, M. I. Ibrahem, M. Mahmoud, W. Alasmary, M. M. Fouda, K. H. Almotairi, and Z. M. Fadlullah, “Privacy-preserving federated- learning-based net-energy forecasting,” in SoutheastCon 2022, 2022, pp. 133–139

    work page 2022

  6. [6]

    Privacy-preserving and efficient decentralized federated learning-based energy theft detector,

    M. I. Ibrahem, M. Mahmoud, M. M. Fouda, B. M. ElHalawany, and W. Alasmary, “Privacy-preserving and efficient decentralized federated learning-based energy theft detector,” in GLOBECOM 2022 - 2022 IEEE Global Communications Conference , 2022, pp. 287–292

    work page 2022

  7. [7]

    On COVID-19 prediction using asynchronous federated learning-based agile radiograph screening booths,

    S. Sakib et al., “On COVID-19 prediction using asynchronous federated learning-based agile radiograph screening booths,” in ICC 2021 - IEEE International Conference on Communications , 2021

    work page 2021

  8. [8]

    Asynchronous federated learning-based ECG analysis for arrhythmia detection,

    S. Sakib, M. M. Fouda, Z. M. Fadlullah, K. Abualsaud, E. Yaacoub, and M. Guizani, “Asynchronous federated learning-based ECG analysis for arrhythmia detection,” in 2021 IEEE International Mediterranean Conference on Communications and Networking (MeditCom) , 2021

    work page 2021

Show all 23 references
  1. [9]

    Toward asynchronously weight updating federated learning for AI-on-edge IoT systems,

    Y . Gupta, Z. M. Fadlullah, and M. M. Fouda, “Toward asynchronously weight updating federated learning for AI-on-edge IoT systems,” in 2022 IEEE International Conference on Internet of Things and Intelligence Systems (IoTaIS), 2022

    2022

  2. [10]

    A lightweight federated learning based privacy preserving B5G pandemic response network using unmanned aerial vehicles: A proof-of-concept,

    N. Nasser, Z. M. Fadlullah, M. M. Fouda, A. Ali, and M. Imran, “A lightweight federated learning based privacy preserving B5G pandemic response network using unmanned aerial vehicles: A proof-of-concept,” Computer Networks , vol. 205, article no. 108672, 2022

    2022

  3. [11]

    Adaboost-based security level classification of mobile intelligent terminals,

    F. Wang, D. Jiang, H. Wen et al. , “Adaboost-based security level classification of mobile intelligent terminals,”Journal of Supercomputing, vol. 75, pp. 7460–7478, 2019

    2019

  4. [12]

    Joint provisioning of QoS and se- curity in IoD networks: Classical optimization meets AI,

    Z. M. Fadlullah and A. Benslimane, “Joint provisioning of QoS and se- curity in IoD networks: Classical optimization meets AI,” IEEE Internet of Things Magazine , vol. 4, no. 4, pp. 40–46, 2021

    2021

  5. [13]

    CICIoT2023: A real-time dataset and benchmark for large- scale attacks in IoT environment,

    E. C. P. Neto, S. Dadkhah, R. Ferreira, A. Zohourian, R. Lu, and A. A. Ghorbani, “CICIoT2023: A real-time dataset and benchmark for large- scale attacks in IoT environment,” Sensors, vol. 23, no. 13, article no. 5941, 2023

    2023

  6. [14]

    Data resampling for federated learning with non-IID labels,

    Z. Tang, Z. Hu, S. Shi, Y .-m. Cheung, Y . Jin, Z. Ren, and X. Chu, “Data resampling for federated learning with non-IID labels,” in FTL-IJCAI’21, 2021

    2021

  7. [15]

    AdaBoost-CNN: an adaptive boosting algorithm for convolutional neural networks to classify multi-class imbalanced datasets using transfer learning,

    A. Taherkhani, G. Cosma, and T. McGinnity, “AdaBoost-CNN: an adaptive boosting algorithm for convolutional neural networks to classify multi-class imbalanced datasets using transfer learning,” Neurocomput- ing, vol. 404, pp. 351–366, 2020

    2020

  8. [16]

    LoAdaBoost: loss-based AdaBoost federated machine learning with reduced computa- tional complexity on IID and non-IID intensive care data,

    L. Huang, Y . Yin, Z. Fu, S. Zhang, H. Deng, and D. Liu, “LoAdaBoost: loss-based AdaBoost federated machine learning with reduced computa- tional complexity on IID and non-IID intensive care data,” 2020

    2020

  9. [17]

    Differential privacy for deep and federated learning: A survey,

    A. El Ouadrhiri and A. Abdelhadi, “Differential privacy for deep and federated learning: A survey,” IEEE Access , vol. 10, pp. 22 359–22 380, 2022

    2022

  10. [18]

    Federated learning with differential privacy: Algorithms and performance analysis,

    K. Wei, J. Li, M. Ding, C. Ma, H. H. Yang, F. Farokhi, S. Jin, T. Q. S. Quek, and H. Vincent Poor, “Federated learning with differential privacy: Algorithms and performance analysis,” IEEE Transactions on Information F orensics and Security, vol. 15, pp. 3454–3469, 2020

    2020

  11. [19]

    Evaluating differentially private machine learning in practice,

    B. Jayaraman and D. Evans, “Evaluating differentially private machine learning in practice,” in Proceedings of the 28th USENIX Conference on Security Symposium , 2019, p. 1895–1912

    2019

  12. [20]

    Communication and computation efficiency in federated learning: A survey,

    O. R. A. Almanifi, C.-O. Chow, M.-L. Tham, J. H. Chuah, and J. Kane- san, “Communication and computation efficiency in federated learning: A survey,” Internet of Things , vol. 22, p. 100742, 2023

    2023

  13. [21]

    Communication-efficient federated learning via quantized compressed sensing,

    Y . Oh, N. Lee, Y .-S. Jeon, and H. V . Poor, “Communication-efficient federated learning via quantized compressed sensing,” IEEE Transactions on Wireless Communications , vol. 22, no. 2, pp. 1087–1100, 2022

    2022

  14. [22]

    A robust federated learning approach for combating attacks against IoT systems under non-IID challenges,

    E. Gad, Z. M. Fadlullah, and M. M. Fouda, “A robust federated learning approach for combating attacks against IoT systems under non-IID challenges,” in 2024 International Conference on Smart Applications, Communications and Networking (SmartNets) , 2024

    2024

  15. [23]

    Combating IoT attacks in AI-driven networks via robust and resource-efficient federated learning,

    E. Gad, “Combating IoT attacks in AI-driven networks via robust and resource-efficient federated learning,” Master’s thesis, Western University, 2024. [Online]. Available: https://ir.lib.uwo.ca/etd/10580/

    2024

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.