pith. sign in

arxiv: 2605.05896 · v1 · submitted 2026-05-07 · 💻 cs.LG · cs.AI

VARS-FL: Validation-Aligned Client Selection for Non-IID Federated Learning in IoT Systems

Mohamed Lakas , Mohamed Amine Ferrag This is my paper

Pith reviewed 2026-05-08 15:37 UTC · model grok-4.3

classification 💻 cs.LG cs.AI
keywords federated learningclient selectionnon-IID dataIoTintrusion detectionreputation scoringvalidation lossEdge-IIoTset
0
0 comments X

The pith

Client selection scored by server validation loss reduction speeds convergence in non-IID IoT federated learning.

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

The paper proposes VARS-FL to fix slow and unstable training in federated learning when data on different devices follows different distributions. It replaces stateless or locally measured client selection with scores based on how much each client's update lowers the loss measured on a server-held validation set. These signals are turned into reputation values that average recent contributions inside a sliding window and add a participation adjustment, then used to pick clients for the next round. The approach leaves local training and the central aggregation step unchanged. In practice this produces higher accuracy and fewer rounds to target performance on heterogeneous IoT intrusion data.

Core claim

VARS-FL quantifies each client's contribution by the reduction in server-side validation loss after its update is applied, aggregates these signals into a reputation score via a sliding-window average plus a logarithmically scaled participation term, and selects clients for each round according to that reputation. The method requires no modification to local training or to FedAvg aggregation. On a 15-class non-IID intrusion detection task drawn from the Edge-IIoTset dataset with 100 clients, it improves final accuracy and F1-Macro, reduces loss, and reaches 80 percent accuracy in up to 36 percent fewer rounds than FedAvg, Oort, or Power-of-Choice across multiple random seeds.

What carries the argument

Validation-aligned reputation scoring, which converts the per-round drop in server validation loss into a history-aware client ranking for selection decisions.

If this is right

  • Target accuracy is reached with up to 36 percent fewer communication rounds.
  • Final accuracy and F1-Macro scores rise compared with random selection and two published alternatives.
  • Training stability improves under the heterogeneous data patterns found across IoT devices.
  • No changes to client-side optimization or central aggregation are required.
  • Gains appear consistently across repeated trials with different random seeds.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same validation-reduction signal could be tested in other non-IID domains where a central validation set is feasible, such as cross-hospital medical imaging.
  • Fewer total rounds would lower cumulative communication and energy costs for battery-powered IoT fleets.
  • If the server validation set drifts over time, a lightweight refresh schedule might be needed to keep the reputation scores aligned.
  • Combining the reputation term with an explicit measure of client data diversity could handle more extreme distribution shifts.

Load-bearing premise

That the reduction in server validation loss caused by a client's update is a reliable and unbiased indicator of its value to the global model, and that the validation set stays representative even though client data distributions differ.

What would settle it

Running the identical 100-client non-IID Edge-IIoTset experiment and finding no measurable gains in accuracy, F1-Macro, or rounds needed to reach 80 percent accuracy for VARS-FL versus the baselines would falsify the performance claims.

Figures

Figures reproduced from arXiv: 2605.05896 by Mohamed Amine Ferrag, Mohamed Lakas.

Figure 1
Figure 1. Figure 1: Non-IID class presence across 100 clients (seed 42). Each row is a client sorted by class count; each column is an attack class. White indicates absence. Dataset sizes range from 426 to 5,152 samples per client (mean: 3,250); the number of local classes ranges from 1 to 15 (mean: 7.9), reflecting the heterogeneous data distributions characteristic of real IoT deployments. limitations through a unified fram… view at source ↗
Figure 2
Figure 2. Figure 2: Overview of VARS-FL, a validation-aligned, reputation-based client selection framework for federated learning. In each communication round, the server broadcasts the global model to selected clients, which perform local training on heterogeneous (non-IID) data and return updated models. The server evaluates each client’s contribution using validation-loss improvement, producing a globally aligned quality s… view at source ↗
Figure 3
Figure 3. Figure 3: Architecture of the fully connected DNN used in the federated learning experiments. The model takes 43 standardized numeric features as input, followed by three hidden layers of sizes 128, 64, and 32 with ReLU activations. Dropout (p = 0.3) is applied after the first two hidden layers, and the output layer uses softmax activation for multi-class classification. where z1 = Dropout(h1, p = 0.3) and z2 = Drop… view at source ↗
Figure 4
Figure 4. Figure 4: Class distribution of the training split (515,232 samples) after view at source ↗
Figure 4
Figure 4. Figure 4: B. Models and Configurations We compare VARS-FL against three representative client selection baselines: FedAvg [1], Oort [21], and Power-of￾Choice [24]. To ensure a fair comparison, all methods use the same model architecture, optimizer, learning rate, number of local epochs, and aggregation rule (FedAvg). This isolates the effect of client selection from other factors. The experimental configuration is s… view at source ↗
Figure 5
Figure 5. Figure 5: Test accuracy of all strategies over 100 rounds (100 clients, seed 123) view at source ↗
Figure 6
Figure 6. Figure 6: Test loss of all strategies over 100 rounds (100 clients, seed 123) view at source ↗
Figure 7
Figure 7. Figure 7: Per round accuracy delta relative to FedAvg for Oort, PoC, and VARS view at source ↗
Figure 8
Figure 8. Figure 8: Class data proportions for the stratified (n = 110,407) and uniform (n = 2,250) validation sets. The stratified set reflects the natural class imbalance of Edge-IIoTset. The uniform set assigns exactly 150 samples per class and aligns with the perfect-uniform baseline (dashed line at 6.7%). in Table VIII. The table shows the test-set performance (mean ± std over three seeds) under both validation set distr… view at source ↗
read the original abstract

Federated learning (FL) systems typically employ stateless client selection, treating each communication round independently and ignoring accumulated evidence of client contribution quality. Under non-IID data, this leads to slow convergence and unstable training, particularly when selection relies on local proxies (e.g., training loss) that are misaligned with the global optimization objective. These challenges are especially pronounced in Internet of Things (IoT) and Industrial IoT (IIoT) environments, where data is highly heterogeneous and distributed across devices observing different traffic patterns. In this paper, we propose VARS-FL (Validation-Aligned Reputation Scoring for Federated Learning), a client selection framework that quantifies each client's contribution using the reduction in server-side validation loss induced by its update. These per-round signals are aggregated into a Reputation score that combines a sliding-window average of recent contributions with a logarithmically scaled participation term, enabling robust exploration-exploitation selection. VARS-FL requires no changes to local training or aggregation and remains fully compatible with standard FedAvg. We evaluate VARS-FL on a 15-class non-IID IoT intrusion detection task using the Edge-IIoTset dataset, with 100 clients across multiple seeds, and compare it against FedAvg, Oort, and Power-of-Choice. VARS-FL consistently improves accuracy, F1-Macro, and loss, while accelerating convergence (up to 36% fewer rounds to reach 80% accuracy). These results demonstrate that validation-aligned, history-aware client selection provides a more reliable and efficient training process for federated learning in heterogeneous IoT environments.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 2 minor

Summary. The manuscript proposes VARS-FL, a client selection strategy for federated learning in non-IID IoT settings. It computes a reputation score for each client based on the reduction in server-side validation loss caused by incorporating the client's update, combined with a sliding-window average and a logarithmically scaled participation count. This score is used to select clients each round in a manner compatible with FedAvg. The method is evaluated on the Edge-IIoTset dataset for 15-class intrusion detection with 100 clients under non-IID partitioning, showing improvements in accuracy, F1-Macro, and convergence speed (up to 36% fewer rounds) over FedAvg, Oort, and Power-of-Choice across multiple seeds.

Significance. Should the validation-aligned reputation mechanism prove robust, the work offers a practical enhancement to client selection in heterogeneous federated learning without requiring changes to local training procedures. The history-aware aspect addresses limitations of stateless selection, and the empirical gains on a realistic IoT intrusion detection task suggest potential applicability in resource-constrained environments. The use of multiple seeds and standard metrics strengthens the empirical case.

major comments (3)
  1. Abstract and §4 (Evaluation): The server-side validation set used to compute loss reductions is never described—its size, construction (e.g., how the 15 classes are sampled), class balance, or refresh policy across rounds are omitted. This detail is load-bearing for the central claim, because the reputation score and subsequent selection rule rest on the assumption that validation-loss reduction is an unbiased proxy for contribution to the global objective under 15-class non-IID partitions; without it, the reported gains cannot be verified as robust to underrepresented traffic patterns or devices.
  2. §3 (Reputation Score Definition): The reputation score depends on two free parameters (sliding-window size and the logarithmic scaling coefficient for participation count) whose specific values and sensitivity are not reported or ablated. Because the selection rule is defined directly in terms of these parameters, the 36% convergence improvement and accuracy/F1 gains may be artifacts of particular tuning rather than a general property of validation-aligned scoring.
  3. §4 (Experiments): No statistical testing (confidence intervals, p-values, or variance across the multiple seeds) is provided for the accuracy, F1-Macro, or round-to-80%-accuracy metrics, nor is there analysis of robustness to alternative non-IID partitionings. This weakens the claim that VARS-FL “consistently improves” performance relative to the three baselines.
minor comments (2)
  1. §2 (Related Work): The positioning against Oort and Power-of-Choice is clear, but the manuscript would benefit from explicit comparison to other recent validation- or loss-based client selection methods that also avoid local proxies.
  2. Notation throughout: The equations for the per-round loss reduction and the aggregated reputation score should be numbered and presented formally rather than described only in prose, to aid reproducibility.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive and detailed feedback. We address each major comment below and will revise the manuscript to incorporate the suggested improvements for clarity, reproducibility, and statistical rigor.

read point-by-point responses

  1. Referee: Abstract and §4 (Evaluation): The server-side validation set used to compute loss reductions is never described—its size, construction (e.g., how the 15 classes are sampled), class balance, or refresh policy across rounds are omitted. This detail is load-bearing for the central claim, because the reputation score and subsequent selection rule rest on the assumption that validation-loss reduction is an unbiased proxy for contribution to the global objective under 15-class non-IID partitions; without it, the reported gains cannot be verified as robust to underrepresented traffic patterns or devices.

    Authors: We agree that explicit details on the server-side validation set are necessary to support the claims and enable verification. In the revised manuscript, we will expand Section 4 to describe the validation set as a fixed, held-out subset comprising approximately 15% of the Edge-IIoTset data, constructed via stratified sampling to ensure representation of all 15 classes with reported class balance statistics, and not refreshed across rounds to maintain consistent signals for the reputation mechanism. revision: yes

  2. Referee: §3 (Reputation Score Definition): The reputation score depends on two free parameters (sliding-window size and the logarithmic scaling coefficient for participation count) whose specific values and sensitivity are not reported or ablated. Because the selection rule is defined directly in terms of these parameters, the 36% convergence improvement and accuracy/F1 gains may be artifacts of particular tuning rather than a general property of validation-aligned scoring.

    Authors: We acknowledge that the specific hyperparameter values and their sensitivity were not sufficiently documented. The revised paper will explicitly report the sliding-window size (W=5) and logarithmic scaling coefficient used in experiments. We will also add an ablation study (in the main text or appendix) varying these parameters over reasonable ranges and demonstrating that the reported gains in convergence speed and accuracy remain consistent, addressing concerns about tuning artifacts. revision: yes

  3. Referee: §4 (Experiments): No statistical testing (confidence intervals, p-values, or variance across the multiple seeds) is provided for the accuracy, F1-Macro, or round-to-80%-accuracy metrics, nor is there analysis of robustness to alternative non-IID partitionings. This weakens the claim that VARS-FL “consistently improves” performance relative to the three baselines.

    Authors: We agree that adding statistical details would strengthen the empirical claims. The revision will include standard deviations and confidence intervals for accuracy, F1-Macro, and convergence metrics across the reported seeds. For robustness to alternative non-IID partitionings, we will add a short analysis using at least one additional partitioning strategy (e.g., varying Dirichlet alpha) in the appendix, while noting that the primary Edge-IIoTset setup is representative of the target IoT scenario; full exhaustive testing may be limited by space but the added results will support the consistency argument. revision: partial

Circularity Check

0 steps flagged

VARS-FL reputation scoring is an explicit heuristic with no reduction to inputs by construction

full rationale

The paper defines client reputation directly from per-round reductions in server-side validation loss (aggregated via sliding-window average plus log participation) and uses this for selection. This construction is not fitted to the final accuracy or F1 target; the performance gains are asserted via empirical comparison on Edge-IIoTset rather than derived from any self-referential equation. No load-bearing step equates a claimed prediction to its own inputs, no uniqueness theorem is invoked, and no self-citation chain supports the core mechanism. The derivation chain is therefore self-contained and non-circular.

Axiom & Free-Parameter Ledger

2 free parameters · 2 axioms · 1 invented entities

The framework rests on empirical signals from a held-out validation set and two aggregation parameters whose specific values are not stated in the abstract; it assumes standard FedAvg compatibility and that validation loss reduction correlates with global utility.

free parameters (2)
  • sliding-window size
    Controls how many recent rounds contribute to the average reputation; value not provided in abstract.
  • logarithmic scaling coefficient for participation
    Determines the weight of the participation term in the final reputation score; value not provided in abstract.
axioms (2)
  • domain assumption Server maintains a representative validation set whose loss reduction accurately reflects client contribution to the global model.
    Central to the scoring rule and stated as the basis for alignment with the optimization objective.
  • domain assumption VARS-FL requires no changes to local training or the FedAvg aggregation step.
    Explicit compatibility claim in the abstract.
invented entities (1)
  • Reputation score no independent evidence
    purpose: Aggregates per-round validation loss reduction signals with participation history to guide client selection.
    New quantity defined by the paper as the combination of sliding-window average and log-scaled participation term.

pith-pipeline@v0.9.0 · 5590 in / 1589 out tokens · 61492 ms · 2026-05-08T15:37:35.468722+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

44 extracted references · 44 canonical work pages

  1. [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,” inProceedings of the 20th International Conference on Artificial Intelligence and Statistics (AISTATS), 2017, pp. 1273–1282

    work page 2017

  2. [2]

    Advances and open problems in federated learning,

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

    work page 2021

  3. [3]

    Edge-iiotset: A new comprehensive realistic cyber security dataset of iot and iiot applications for centralized and federated learning,

    M. A. Ferrag, O. Friha, D. Hamouda, L. Maglaras, and H. Janicke, “Edge-iiotset: A new comprehensive realistic cyber security dataset of iot and iiot applications for centralized and federated learning,”IEEE Access, vol. 10, pp. 40 281–40 306, 2022

    work page 2022

  4. [4]

    Decentralized federated learning with non-iid data: Challenges, trends, and future opportunities,

    W.-C. Chung, C.-A. Lo, Y .-H. Lin, Z.-H. Chen, and C.-L. Hung, “Decentralized federated learning with non-iid data: Challenges, trends, and future opportunities,”ACM Computing Surveys, vol. 58, no. 8, pp. 1–41, 2026

    work page 2026

  5. [5]

    Ppss: A privacy-preserving secure framework using blockchain-enabled feder- ated deep learning for industrial iots,

    D. Hamouda, M. A. Ferrag, N. Benhamida, and H. Seridi, “Ppss: A privacy-preserving secure framework using blockchain-enabled feder- ated deep learning for industrial iots,”Pervasive and Mobile Computing, vol. 88, p. 101738, 2023

    work page 2023

  6. [6]

    Revolutionizing intrusion detection in industrial iot with distributed learning and deep generative techniques,

    D. Hamouda, M. A. Ferrag, N. Benhamida, H. Seridi, and M. C. Ghanem, “Revolutionizing intrusion detection in industrial iot with distributed learning and deep generative techniques,”Internet of Things, vol. 26, p. 101149, 2024

    work page 2024

  7. [7]

    Federated learning: Strategies for improving communication efficiency,

    J. Kone ˇcný, H. B. McMahan, F. X. Yu, P. Richtárik, A. T. Suresh, and D. Bacon, “Federated learning: Strategies for improving communication efficiency,” inNIPS Workshop on Private Multi-Party Machine Learning, 2016

    work page 2016

  8. [8]

    Practical secure aggregation for privacy-preserving machine learning,

    K. Bonawitz, V . Ivanov, B. Kreuter, A. Marcedone, H. B. McMahan, S. Patel, D. Ramage, A. Segal, and K. Seth, “Practical secure aggregation for privacy-preserving machine learning,” inProceedings of the ACM SIGSAC Conference on Computer and Communications Security (CCS), 2017

    work page 2017

  9. [9]

    QSGD: Communication-efficient SGD via gradient quantization and encoding,

    D. Alistarh, J. Grubic, J. Li, R. Tomioka, and M. V ojnovic, “QSGD: Communication-efficient SGD via gradient quantization and encoding,” inAdvances in Neural Information Processing Systems (NeurIPS), 2017

    work page 2017

  10. [10]

    Sparse communication for distributed gra- dient descent,

    A. F. Aji and K. Heafield, “Sparse communication for distributed gra- dient descent,” inProceedings of the Conference on Empirical Methods in Natural Language Processing (EMNLP), 2017

    work page 2017

  11. [11]

    Client selection for federated learning with heterogeneous resources in mobile edge,

    T. Nishio and R. Yonetani, “Client selection for federated learning with heterogeneous resources in mobile edge,” inProceedings of the IEEE International Conference on Communications (ICC), 2019, pp. 1–7

    work page 2019

  12. [12]

    Joint device scheduling and resource allocation for latency constrained federated learning,

    W. Shi, S. Zhou, Z. Niu, M. Jiang, and L. Geng, “Joint device scheduling and resource allocation for latency constrained federated learning,”IEEE Transactions on Wireless Communications, vol. 20, no. 1, pp. 453–467, 2021

    work page 2021

  13. [13]

    Federated learning for cyber physical systems: a comprehensive survey,

    M. K. Quan, P. N. Pathirana, M. Wijayasundara, S. Setunge, D. C. Nguyen, C. G. Brinton, D. J. Love, and H. V . Poor, “Federated learning for cyber physical systems: a comprehensive survey,”IEEE Communications Surveys & Tutorials, 2025

    work page 2025

  14. [14]

    Federated learning for computationally constrained heterogeneous devices: A survey,

    K. Pfeiffer, M. Rapp, R. Khalili, and J. Henkel, “Federated learning for computationally constrained heterogeneous devices: A survey,”ACM Computing Surveys, vol. 55, no. 14s, pp. 1–27, 2023

    work page 2023

  15. [15]

    A systematic literature review of robust federated learning: Issues, solutions, and future research directions,

    M. P. Uddin, Y . Xiang, M. Hasan, J. Bai, Y . Zhao, and L. Gao, “A systematic literature review of robust federated learning: Issues, solutions, and future research directions,”ACM Computing Surveys, vol. 57, no. 10, pp. 1–62, 2025

    work page 2025

  16. [16]

    Client selection in federated learning: Principles, challenges, and opportunities,

    L. Fu, H. Zhang, G. Gao, M. Zhang, and X. Liu, “Client selection in federated learning: Principles, challenges, and opportunities,”IEEE Internet of Things Journal, vol. 10, no. 24, pp. 21 811–21 819, 2023

    work page 2023

  17. [17]

    Green federated learning: A new era of green aware ai,

    D. Thakur, A. Guzzo, G. Fortino, and F. Piccialli, “Green federated learning: A new era of green aware ai,”ACM Computing Surveys, vol. 57, no. 8, pp. 1–36, 2025

    work page 2025

  18. [18]

    Bias in federated learning: A compre- hensive survey,

    N. Benarba and S. Bouchenak, “Bias in federated learning: A compre- hensive survey,”ACM Computing Surveys, vol. 57, no. 11, pp. 1–36, 2025

    work page 2025

  19. [19]

    Heterogeneous feder- ated learning: State-of-the-art and research challenges,

    M. Ye, X. Fang, B. Du, P. C. Yuen, and D. Tao, “Heterogeneous feder- ated learning: State-of-the-art and research challenges,”ACM Computing Surveys, vol. 56, no. 3, pp. 1–44, 2023

    work page 2023

  20. [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. Kanesan, “Communication and computation efficiency in federated learning: A survey,”Internet of Things, vol. 22, p. 100742, 2023

    work page 2023

  21. [21]

    Oort: Efficient federated learning via guided participant selection,

    F. Lai, X. Zhu, H. V . Madhyastha, and M. Chowdhury, “Oort: Efficient federated learning via guided participant selection,” inProceedings of the 15th USENIX Symposium on Operating Systems Design and Implementation (OSDI), 2021, pp. 19–35

    work page 2021

  22. [22]

    Decentralized federated learning: Fundamentals, state of the art, frameworks, trends, and challenges,

    E. T. M. Beltrán, M. Q. Pérez, P. M. S. Sánchez, S. L. Bernal, G. Bovet, M. G. Pérez, G. M. Pérez, and A. H. Celdrán, “Decentralized federated learning: Fundamentals, state of the art, frameworks, trends, and challenges,”IEEE Communications Surveys & Tutorials, vol. 25, no. 4, pp. 2983–3013, 2023

    work page 2023

  23. [23]

    Federated optimization in heterogeneous networks,

    T. Li, A. K. Sahu, M. Zaheer, M. Sanjabi, A. Talwalkar, and V . Smith, “Federated optimization in heterogeneous networks,” inProceedings of Machine Learning and Systems (MLSys), 2020

    work page 2020

  24. [24]

    Towards understanding biased client selection in federated learning,

    Y . Jee Cho, J. Wang, and G. Joshi, “Towards understanding biased client selection in federated learning,” inProceedings of The 25th International Conference on Artificial Intelligence and Statistics, ser. Proceedings of Machine Learning Research, vol. 151. PMLR, 2022, pp. 10 351–10 375

    work page 2022

  25. [25]

    Active federated learning,

    J. Goetz, K. Malik, D. Bui, S. Moon, H. Liu, and A. Kumar, “Active federated learning,”arXiv preprint arXiv:1909.12641, 2019

    work page arXiv 1909

  26. [26]

    Fed- Cor: Correlation-based active client selection strategy for heterogeneous federated learning,

    M. Tang, X. Ning, Y . Wang, J. Sun, Y . Wang, H. Li, and C. Yao, “Fed- Cor: Correlation-based active client selection strategy for heterogeneous federated learning,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2022, pp. 10 102– 10 111

    work page 2022

  27. [27]

    FLTrust: Byzantine- robust federated learning via trust bootstrapping,

    X. Cao, M. Fang, J. Liu, and N. Z. Gong, “FLTrust: Byzantine- robust federated learning via trust bootstrapping,” inProceedings of the Network and Distributed System Security Symposium (NDSS), 2021

    work page 2021

  28. [28]

    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,” inInternational Conference on Learning Representations (ICLR), 2020

    work page 2020

  29. [29]

    SCAFFOLD: Stochastic controlled averaging for federated learning,

    S. P. Karimireddy, S. Kale, M. Mohri, S. J. Reddi, S. Stich, and A. T. Suresh, “SCAFFOLD: Stochastic controlled averaging for federated learning,” inProceedings of the International Conference on Machine Learning (ICML), 2020, pp. 5132–5143

    work page 2020

  30. [30]

    Model-contrastive federated learning,

    Q. Li, B. He, and D. Song, “Model-contrastive federated learning,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2021, pp. 10 713–10 722

    work page 2021

  31. [31]

    FedScale: Benchmarking model and system perfor- mance of federated learning at scale,

    F. Lai, Y . Dai, S. Singapuram, J. Liu, X. Zhu, H. V . Madhyastha, and M. Chowdhury, “FedScale: Benchmarking model and system perfor- mance of federated learning at scale,” inProceedings of the International Conference on Machine Learning (ICML), 2022, pp. 11 814–11 827

    work page 2022

  32. [32]

    Machine learning with adversaries: Byzantine tolerant gradient descent,

    P. Blanchard, E. M. El Mhamdi, R. Guerraoui, and J. Stainer, “Machine learning with adversaries: Byzantine tolerant gradient descent,” inAd- vances in Neural Information Processing Systems (NeurIPS), 2017, pp. 119–129

    work page 2017

  33. [33]

    The hidden vul- nerability of distributed learning in Byzantium,

    E. M. El Mhamdi, R. Guerraoui, and S. Rouault, “The hidden vul- nerability of distributed learning in Byzantium,” inProceedings of the International Conference on Machine Learning (ICML), 2018, pp. 3521– 3530

    work page 2018

  34. [34]

    Exploiting shared representations for personalized federated learning,

    L. Collins, H. Hassani, A. Mokhtari, and S. Shakkottai, “Exploiting shared representations for personalized federated learning,” inProceed- ings of the International Conference on Machine Learning (ICML), 2021, pp. 2089–2099

    work page 2021

  35. [35]

    Personalized federated learning with Moreau envelopes,

    C. T. Dinh, N. H. Tran, and T. D. Nguyen, “Personalized federated learning with Moreau envelopes,” inAdvances in Neural Information Processing Systems (NeurIPS), 2020, pp. 21 394–21 405

    work page 2020

  36. [36]

    On upper-confidence bound policies for switching bandit problems,

    A. Garivier and É. Moulines, “On upper-confidence bound policies for switching bandit problems,” inProceedings of the International Conference on Algorithmic Learning Theory (ALT), 2011, pp. 174–188

    work page 2011

  37. [37]

    A tutorial on Thompson sampling,

    D. J. Russo, B. Van Roy, A. Kazerouni, I. Osband, and Z. Wen, “A tutorial on Thompson sampling,”Foundations and Trends in Machine Learning, vol. 11, no. 1, pp. 1–96, 2018

    work page 2018

  38. [38]

    A survey on security and privacy of federated learning,

    V . Mothukuri, R. M. Parizi, S. Pouriyeh, Y . Huang, A. Dehghantanha, and G. Srivastava, “A survey on security and privacy of federated learning,”Future Generation Computer Systems, vol. 115, pp. 619–640, 2021

    work page 2021

  39. [39]

    DÏoT: A federated self-learning anomaly detec- tion system for iot,

    T. D. Nguyen, S. Marchal, M. Miettinen, H. Fereidooni, N. Asokan, and A.-R. Sadeghi, “DÏoT: A federated self-learning anomaly detec- tion system for iot,” inIEEE International Conference on Distributed Computing Systems (ICDCS), 2019

    work page 2019

  40. [40]

    Federated learning for malware detection in IoT devices,

    V . Rey, P. M. Sanchez Sanchez, A. H. Celdrán, and G. Bovet, “Federated learning for malware detection in IoT devices,”Computer Networks, vol. 204, p. 108693, 2022

    work page 2022

  41. [41]

    Internet of things intrusion detection: Centralized, on-device, or feder- ated learning?

    S. T. Rahman, A. Mansoor, M. Shamim Hossain, and M. M. Rahman, “Internet of things intrusion detection: Centralized, on-device, or feder- ated learning?”IEEE Network, vol. 34, no. 6, pp. 310–317, 2020

    work page 2020

  42. [42]

    Toward generating a new intrusion detection dataset and intrusion traffic characterization,

    I. Sharafaldin, A. H. Lashkari, and A. A. Ghorbani, “Toward generating a new intrusion detection dataset and intrusion traffic characterization,” in Proceedings of the 4th International Conference on Information Systems Security and Privacy (ICISSP), 2018, pp. 108–116

    work page 2018

  43. [43]

    A new distributed architecture for evaluating AI-based security systems at the edge: Network TON_IoT datasets,

    N. Moustafa, “A new distributed architecture for evaluating AI-based security systems at the edge: Network TON_IoT datasets,”Sustainable Cities and Society, vol. 72, p. 102994, 2021

    work page 2021

  44. [44]

    N-BaIoT: Network-based detection of IoT botnet attacks using deep autoencoders,

    Y . Meidan, M. Bohadana, Y . Mathov, Y . Mirsky, A. Shabtai, D. Bre- itenbacher, and Y . Elovici, “N-BaIoT: Network-based detection of IoT botnet attacks using deep autoencoders,”IEEE Pervasive Computing, vol. 17, no. 3, pp. 12–22, 2018

    work page 2018