pith. sign in

arxiv: 2606.10595 · v1 · pith:T3WS5LJ3new · submitted 2026-06-09 · 💻 cs.CR · cs.AI

From Data Heterogeneity to Convergence: A Data-Centric Review of Federated Learning

Huong Nguyen , Micka\"el Bettinelli , Amirhossein Ghaffari , Alexandre Benoit , Hong-Tri Nguyen , Susanna Pirttikangas , Lauri Lov\'en This is my paper

Pith reviewed 2026-06-27 12:40 UTC · model grok-4.3

classification 💻 cs.CR cs.AI
keywords federated learningdata heterogeneitynon-IID dataconvergencerobustnessdata splittingadversarial defensessurvey
0
0 comments X

The pith

This survey ranks non-IID data traits by influence on federated learning convergence and makes the convergence-robustness trade-off explicit.

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

The paper establishes a data-centric synthesis of federated learning by breaking non-IID data into measurable traits and ranking them as strong, medium, or light in their effect on convergence speed and stability. It maps common experimental splitting methods to the real-world distributions they emulate, identifies the artifacts they introduce, and shows how those artifacts change target accuracy. The work also evaluates data-related vulnerabilities and defenses under both clean and adversarial conditions to clarify performance trade-offs. These elements together aim to give practitioners concrete guidance for designing systems that achieve more stable and predictable training outcomes.

Core claim

The survey claims to deliver the first complete understanding of data-related challenges that govern federated learning by analyzing non-IID traits and ranking their convergence influence, connecting splitting practices to real phenomena while exposing artifacts, and reporting how defenses affect convergence and robustness in clean versus adversarial settings, with clear takeaways for each concern.

What carries the argument

The ranking of non-IID traits by convergence influence (strong, medium, light) together with the explicit mapping of splitting practices to real phenomena and the convergence-robustness trade-off analysis.

If this is right

  • Practitioners gain prioritized guidance on which data traits to address first to improve convergence speed and stability.
  • Data splitting protocols can be selected to reduce artifacts and better align experimental accuracy with target real-world performance.
  • Defenses can be evaluated and chosen with explicit knowledge of their effects on convergence under both clean and attack conditions.
  • System designs can incorporate the ranked traits and trade-off information to achieve more predictable federated training results.

Where Pith is reading between the lines

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

  • The trait-ranking approach could be tested as a diagnostic tool when deploying federated systems on new data modalities beyond the surveyed domains.
  • Reconciling results across data types suggests the value of developing standardized heterogeneity metrics that generalize beyond images, text, and graphs.
  • The explicit convergence-robustness mapping points toward hybrid defense strategies that optimize both objectives simultaneously rather than treating them separately.
  • Connections drawn between splitting artifacts and accuracy could motivate creation of benchmark datasets that reduce emulation gaps.

Load-bearing premise

The selected literature and experimental evidence across images, texts, and graphs are representative enough to support general rankings of trait influence and to reconcile conflicting results without systematic omission of counter-evidence.

What would settle it

A new review or set of experiments that produces materially different rankings of non-IID trait influences or reveals unaddressed conflicts across data types would falsify the claimed synthesis.

Figures

Figures reproduced from arXiv: 2606.10595 by Alexandre Benoit, Amirhossein Ghaffari, Hong-Tri Nguyen, Huong Nguyen, Lauri Lov\'en, Micka\"el Bettinelli, Susanna Pirttikangas.

Figure 1
Figure 1. Figure 1: Data traits: impact on convergence behaviors (left), and pairwise relations (right). Stronger homophily typically sharpens class separation after message passing, which raises effective ICS, and it can damp local within-class noise, which lowers ICV. Data volume (#samples) primarily improves estimation rather than changing the intrinsic difficulty. With more samples, centroids and covariances are estimated… view at source ↗
Figure 2
Figure 2. Figure 2: Dirichlet-distribution-based splitting Let’s assume there are N clients in the system and C classes in the dataset. For each class c ∈ C, we randomly draw or sample the probability distribution of that class across clients. For example, in the case N=5, this sampled vector can be [0.15, 0.52, 0.1, 0.2, 0.03], meaning that 15% of the data goes to client 1, 52% goes to client 2, and so on. Accordingly, clien… view at source ↗
Figure 3
Figure 3. Figure 3 [PITH_FULL_IMAGE:figures/full_fig_p020_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Cluster-based splitting 21 [PITH_FULL_IMAGE:figures/full_fig_p021_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Temporal-based splitting 4.2.5. Common splitting practices in prior works These data splitting strategies are often constraints imposed by an industrial context. Therefore, mak￾ing synthetic real-world data distributions implies first understanding this context and then splitting the data accordingly. This captures seasonality, concept drift, and availability patterns materially affecting 22 [PITH_FULL_IM… view at source ↗
Figure 6
Figure 6. Figure 6: 2D scattered accuracy by split and datasets. CNN LeNetMLP MobileNet ResNet18 ResNet20 ResNet34 ResNet50 ResNet9 UNet VGG16 ViT-B/16 ('CIFAR-10', 'Dir(0.1 0.3)') ('CIFAR-10', 'Dir(0.3 1)') ('CIFAR-10', 'Dir(1 100)') ('CIFAR-10', 'Dir( 0.1)') ('CIFAR-10', 'Dir( 100)') ('CIFAR-10', 'Feature Shift') ('CIFAR-10', 'IID') ('CIFAR-10', 'Path(1/2)') ('CIFAR-10', 'Path(10)') ('CIFAR-10', 'Path(3/4/5)') ('CIFAR-100',… view at source ↗
Figure 7
Figure 7. Figure 7: Accuracy heatmap Across split, datasets, and models. 29 [PITH_FULL_IMAGE:figures/full_fig_p029_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: 3D scattered accuracy across split, datasets, and models. 4.3. Assisting convergence in non-IID distributions Performance and convergence typically deteriorate as data becomes more non-IID across clients [130, 131], underscoring the importance of carefully choosing and tuning these splitting strategies. To meet this challenge, several approaches have emerged. The first solution focuses on the data itself. … view at source ↗
Figure 9
Figure 9. Figure 9: Label-flipping attack e.g., a small noise or a small patch of image to add over the original image. Consequently, this "sneaky" attack is harder to detect, compared to label flipping, since it functions well under normal conditions and only be triggered under specific malicious input. Interestingly, Bhagoji et al. [152] offer a counterargument to this claim by quantitatively demonstrating that this form of… view at source ↗
Figure 10
Figure 10. Figure 10: Backdoor attack Beyond these two core methods, a stream of poisoning attacks continues to emerge. Most recently, Kasyap and Tripathy in [153] generate adversarial samples using hyperdimensional computing, which projects input data into a high-dimensional space (thousands to millions of dimensions), making it easier to manipulate the data. The key idea of this method is to generate perturbations that shift… view at source ↗
Figure 11
Figure 11. Figure 11: Evasion attack Despite not being as widely discussed as data poisoning, this type of attack still gained significant attention from researchers to propose defense strategies in recent years [172–175]. Among these three methods, Pelta does not report the clean utility, even though it achieves a near-perfect robust accuracy under Self-Attention Gradient Attack (SAGA). This makes it less reliable when the tr… view at source ↗
Figure 12
Figure 12. Figure 12: White-box membership inference attack. The check (✓) indicates that the data was in the training dataset, and the cross (✗) indicates that the data was not. Across membership-inference defenses, the dominant pattern seems substantial privacy gains at negli￾gible clean-utility cost, with a few methods even improving task accuracy. Regularization and calibration approaches that directly target overconfidenc… view at source ↗
Figure 13
Figure 13. Figure 13: Black-box membership inference attack. The check (✓) indicates that the data was in the training dataset, and the cross (✗) indicates that the data was not. magnitude must be carefully calibrated to avoid convergence and utility penalties. Like other attacks, researchers in this community continuously proposed sophisticated techniques for membership inference, which enhance the effectiveness of these atta… view at source ↗
Figure 14
Figure 14. Figure 14: Model starts to forget learned data over time. The confusion matrix at time T+1 only shows good performance on the trained data (5 to 9), not the previously trained ones (0-4) at time T. Among the three common solutions mitigating catastrophic forgetting (replay-based methods [197, 198], regularization-based methods [199, 200], and distillation-based methods [201–203]), memory replay is the one that intro… view at source ↗
read the original abstract

Federated Learning (FL) has emerged as a promising solution for data hunger in centralized learning. This paradigm enables privacy with multiple clients to train a shared-task model collaboratively without exposing their local data. While being a key component in any learning system, data is also a primary source of vulnerabilities and challenges, and a major determinant of a stable and well-converged training. Existing FL reviews describe general foundations, security practices, opportunities, challenges, and applications, without delving into diverse aspects of data and considering problems from the data perspective. They rarely provide a data-lens synthesis that links concrete data properties, split protocols, and defenses to convergence speed and stability. This survey fills that gap with three advances. First, we analyze non-IID into measurable traits and rank their influence on convergence as strong, medium, or light, explaining the mechanisms behind each and reconciling evidence across images, texts, and graphs. Second, we connect experimental splitting practices to the real phenomena they emulate, expose the artifacts they introduce, and show how those artifacts affect target accuracy. Third, we analyze how data-related vulnerabilities and their proposed defenses affect convergence, reporting performance under clean and adversarial conditions to make the convergence-robustness trade-off explicit. To our knowledge, this is the first survey to provide a complete understanding of data-related challenges that govern FL. With clear takeaways distilled for each concern, our work serves as actionable guidance, helping practitioners design their system with predictable convergence and stability.

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

2 major / 1 minor

Summary. The paper is a survey claiming to be the first data-centric review of Federated Learning that fully addresses data-related challenges governing convergence. It advances three contributions: (1) decomposing non-IID into measurable traits and ranking their influence on convergence (strong/medium/light) while explaining mechanisms and reconciling evidence across images, texts, and graphs; (2) mapping experimental data-splitting practices to the real-world phenomena they emulate, exposing introduced artifacts, and showing effects on target accuracy; (3) analyzing data vulnerabilities and defenses under both clean and adversarial conditions to make the convergence-robustness trade-off explicit. The work distills actionable takeaways for practitioners.

Significance. If the literature synthesis and rankings are comprehensive and reproducible, the survey would provide useful guidance by making explicit how specific data properties and defenses affect FL convergence and stability. Credit is due for the multi-modality coverage and for attempting to link splitting artifacts and adversarial settings to convergence outcomes. As a synthesis without new derivations or primary experiments, its value hinges on transparent selection and accurate reconciliation of prior results.

major comments (2)
  1. [Abstract and Introduction] Abstract and Introduction: The central claims of providing a 'complete understanding' and ranking non-IID traits by influence rest on the representativeness of the cited studies. No systematic review protocol, search strategy, inclusion/exclusion criteria, or time frame is described, so the qualitative reconciliation of conflicting results across modalities cannot be verified for completeness or bias. This is load-bearing for the ranking taxonomy and the 'first such survey' assertion.
  2. [Trait-ranking section] Trait-ranking section (the section presenting the strong/medium/light classification): The rankings are derived from narrative synthesis of selected papers rather than quantitative aggregation such as effect-size meta-analysis. Without an explicit protocol, omitted counter-evidence (e.g., recent graph-FL or transformer-client studies) could alter the relative influence ordering, undermining the general claim.
minor comments (1)
  1. [Abstract] Abstract: The claim of being the 'first' survey would benefit from a brief comparison table or sentence distinguishing it from the 'existing FL reviews' mentioned, to clarify novelty without relying solely on author knowledge.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback highlighting the need for greater transparency in our literature synthesis. We agree that explicitly documenting the review process will strengthen the verifiability of the trait rankings and the 'first such survey' claim. We will add a dedicated 'Review Methodology' subsection (approximately 400 words) in the Introduction that details search strategy, databases, time frame, inclusion/exclusion criteria, and reconciliation approach. This revision directly addresses both major comments without altering the core contributions. Point-by-point responses follow.

read point-by-point responses

  1. Referee: [Abstract and Introduction] Abstract and Introduction: The central claims of providing a 'complete understanding' and ranking non-IID traits by influence rest on the representativeness of the cited studies. No systematic review protocol, search strategy, inclusion/exclusion criteria, or time frame is described, so the qualitative reconciliation of conflicting results across modalities cannot be verified for completeness or bias. This is load-bearing for the ranking taxonomy and the 'first such survey' assertion.

    Authors: We acknowledge the absence of an explicit protocol description, which limits independent verification of completeness. In revision we will insert a new 'Review Methodology' subsection that specifies: (i) search terms and databases (Google Scholar, arXiv, IEEE Xplore with keywords 'federated learning non-IID data heterogeneity convergence' 2017–2024); (ii) inclusion criteria (empirical studies reporting convergence metrics under controlled data traits across vision, language, and graph modalities); (iii) exclusion criteria (purely theoretical works without empirical validation, non-peer-reviewed preprints after 2023); and (iv) the narrative reconciliation process used to resolve conflicting modality-specific findings. We will also qualify the 'complete understanding' phrasing to 'comprehensive synthesis of key empirical evidence' to avoid overstatement. These changes make the synthesis reproducible while preserving the multi-modality scope. revision: yes

  2. Referee: [Trait-ranking section] Trait-ranking section (the section presenting the strong/medium/light classification): The rankings are derived from narrative synthesis of selected papers rather than quantitative aggregation such as effect-size meta-analysis. Without an explicit protocol, omitted counter-evidence (e.g., recent graph-FL or transformer-client studies) could alter the relative influence ordering, undermining the general claim.

    Authors: The rankings are intentionally qualitative because standardized effect sizes are unavailable across the heterogeneous experimental setups in the cited literature; a formal meta-analysis would require re-implementation of dozens of studies, which exceeds survey scope. We will revise the section to: (a) explicitly state the narrative synthesis protocol (same as the new methodology subsection); (b) add a limitations paragraph acknowledging that recent graph-FL and transformer-client papers (e.g., post-2023 works) were reviewed but did not overturn the ordering; and (c) include a sensitivity note that future quantitative aggregation could refine the strong/medium/light labels. This maintains the contribution while addressing reproducibility concerns. revision: partial

Circularity Check

0 steps flagged

No circularity: literature synthesis without derivations or self-referential reductions

full rationale

This survey synthesizes external literature on FL data heterogeneity without introducing equations, fitted parameters, predictions, or ansatzes. Claims such as trait rankings and convergence-robustness trade-offs are presented as distillations from reviewed papers rather than derived internally. No self-citation chain is load-bearing for a mathematical result, and no step reduces by construction to the paper's own inputs. The work is self-contained as a review against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

As a survey the paper introduces no new free parameters, axioms, or invented entities; its claims rest on the completeness of the reviewed literature and on the assumption that the cited experimental protocols are faithfully summarized.

pith-pipeline@v0.9.1-grok · 5822 in / 1087 out tokens · 16836 ms · 2026-06-27T12:40:33.186159+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

217 extracted references · 10 canonical work pages

  1. [1]

    C. Yan, B. Gong, Y. Wei, Y. Gao, Deep multi-view enhancement hashing for image retrieval, IEEE Transactions on Pattern Analysis and Machine Intelligence 43 (4) (2020) 1445–1451

    2020

  2. [2]

    C. Yan, Z. Li, Y. Zhang, Y. Liu, X. Ji, Y. Zhang, Depth image denoising using nuclear norm and learn- ing graph model, ACM Transactions on Multimedia Computing, Communications, and Applications (TOMM) 16 (4) (2020) 1–17

    2020

  3. [3]

    McMahan, E

    B. McMahan, E. Moore, D. Ramage, S. Hampson, B. A. y Arcas, Communication-efficient learning of deep networks from decentralized data, in: Artificial intelligence and statistics, PMLR, 2017, pp. 1273–1282

    2017

  4. [4]

    D. C. Nguyen, M. Ding, P. N. Pathirana, A. Seneviratne, J. Li, H. V. Poor, Federated learning for internet of things: A comprehensive survey, IEEE Communications Surveys & Tutorials 23 (3) (2021) 1622–1658

    2021

  5. [5]

    Zhang, Y

    C. Zhang, Y. Xie, H. Bai, B. Yu, W. Li, Y. Gao, A survey on federated learning, Knowledge-Based Systems 216 (2021) 106775

    2021

  6. [6]

    J. Wen, Z. Zhang, Y. Lan, Z. Cui, J. Cai, W. Zhang, A survey on federated learning: challenges and applications, International Journal of Machine Learning and Cybernetics 14 (2) (2023) 513–535. 46

    2023

  7. [7]

    J. Liu, J. Huang, Y. Zhou, X. Li, S. Ji, H. Xiong, D. Dou, From distributed machine learning to federated learning: A survey, Knowledge and Information Systems 64 (4) (2022) 885–917

    2022

  8. [8]

    Aledhari, R

    M. Aledhari, R. Razzak, R. M. Parizi, F. Saeed, Federated learning: A survey on enabling technologies, protocols, and applications, IEEE Access 8 (2020) 140699–140725

    2020

  9. [9]

    AbdulRahman, H

    S. AbdulRahman, H. Tout, H. Ould-Slimane, A. Mourad, C. Talhi, M. Guizani, A survey on federated learning: The journey from centralized to distributed on-site learning and beyond, IEEE Internet of Things Journal 8 (7) (2020) 5476–5497

    2020

  10. [10]

    H. Zhu, J. Xu, S. Liu, Y. Jin, Federated learning on non-iid data: A survey, Neurocomputing 465 (2021) 371–390

    2021

  11. [11]

    Mothukuri, R

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

    2021

  12. [12]

    Hestness, S

    J. Hestness, S. Narang, N. Ardalani, G. Diamos, H. Jun, H. Kianinejad, M. M. A. Patwary, Y. Yang, Y. Zhou, Deep learning scaling is predictable, empirically, arXiv preprint arXiv:1712.00409 (2017)

    Pith/arXiv arXiv 2017

  13. [13]

    C. Sun, A. Shrivastava, S. Singh, A. Gupta, Revisiting unreasonable effectiveness of data in deep learning era, in: Proceedings of the IEEE international conference on computer vision, 2017, pp. 843–852

    2017

  14. [14]

    X. Zhu, C. Vondrick, D. Ramanan, C. C. Fowlkes, Do we need more training data or better models for object detection?., in: BMVC, Vol. 3, Citeseer, 2012

    2012

  15. [15]

    Parsons, F

    Z. Parsons, F. Dou, H. Du, Z. Song, J. Lu, Mobilizing personalized federated learning in infrastructure- less and heterogeneous environments via random walk stochastic admm, Advances in Neural Informa- tion Processing Systems 36 (2023) 36726–36737

    2023

  16. [16]

    Chen, H.-W

    Y.-C. Chen, H.-W. Chen, S.-G. Wang, M.-S. Chen, Space: Single-round participant amalgamation for contribution evaluation in federated learning, Advances in Neural Information Processing Systems 36 (2023) 6422–6441

    2023

  17. [17]

    J. Jia, Z. Yuan, D. Sahabandu, L. Niu, A. Rajabi, B. Ramasubramanian, B. Li, R. Poovendran, Fedgame: a game-theoretic defense against backdoor attacks in federated learning, Advances in Neural Information Processing Systems 36 (2023) 53090–53111

    2023

  18. [18]

    Y. Sun, L. Shen, D. Tao, Understanding how consistency works in federated learning via stage-wise relaxed initialization, Advances in Neural Information Processing Systems 36 (2023) 80543–80574. 47

    2023

  19. [19]

    H. Wang, P. Zheng, X. Han, W. Xu, R. Li, T. Zhang, Fednlr: Federated learning with neuron-wise learning rates, in: Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, 2024, pp. 3069–3080

    2024

  20. [20]

    J. Xu, S. Wan, Y. Li, S. Luo, Z. Chen, Y. Shao, Z. Chen, S.-L. Huang, L. Song, Cooperative multi- modeltrainingforpersonalizedfederatedlearningoverheterogeneousdevices, IEEEJournalofSelected Topics in Signal Processing (2024)

    2024

  21. [21]

    X. Yang, W. Huang, M. Ye, Dynamic personalized federated learning with adaptive differential privacy, Advances in Neural Information Processing Systems 36 (2023) 72181–72192

    2023

  22. [22]

    I. Wang, P. Nair, D. Mahajan, Fluid: Mitigating stragglers in federated learning using invariant dropout, Advances in Neural Information Processing Systems 36 (2023) 73258–73273

    2023

  23. [23]

    Crawshaw, Y

    M. Crawshaw, Y. Bao, M. Liu, Federated learning with client subsampling, data heterogeneity, and un- bounded smoothness: A new algorithm and lower bounds, Advances in Neural Information Processing Systems 36 (2023) 6467–6508

    2023

  24. [24]

    M. Shi, Y. Zhou, K. Wang, H. Zhang, S. Huang, Q. Ye, J. Lv, Prior: Personalized prior for reactivating the information overlooked in federated learning., Advances in Neural Information Processing Systems 36 (2023) 28378–28392

    2023

  25. [25]

    Panchal, S

    K. Panchal, S. Choudhary, S. Mitra, K. Mukherjee, S. Sarkhel, S. Mitra, H. Guan, Flash: concept drift adaptation in federated learning, in: International Conference on Machine Learning, PMLR, 2023, pp. 26931–26962

    2023

  26. [26]

    G. Wan, Z. Shi, W. Huang, G. Zhang, D. Tao, M. Ye, Energy-based backdoor defense against federated graph learning, in: The Thirteenth International Conference on Learning Representations, 2025

    2025

  27. [27]

    S. Kim, Y. Lee, Y. Oh, N. Lee, S. Yun, J. Lee, S. Kim, C. Yang, C. Park, Subgraph federated learning for local generalization, in: The Thirteenth International Conference on Learning Representations, 2025

    2025

  28. [28]

    X. Li, K. Huang, W. Yang, S. Wang, Z. Zhang, On the convergence of fedavg on non-iid data, arXiv preprint arXiv:1907.02189 (2019)

    arXiv 1907

  29. [29]

    T. Li, A. K. Sahu, M. Zaheer, M. Sanjabi, A. Talwalkar, V. Smith, Federated optimization in hetero- geneous networks, Proceedings of Machine learning and systems 2 (2020) 429–450

    2020

  30. [30]

    S. P. Karimireddy, S. Kale, M. Mohri, S. Reddi, S. Stich, A. T. Suresh, Scaffold: Stochastic controlled averaging for federated learning, in: International conference on machine learning, PMLR, 2020, pp. 5132–5143. 48

    2020

  31. [31]

    S. U. Stich, Local sgd converges fast and communicates little, arXiv preprint arXiv:1805.09767 (2018)

    Pith/arXiv arXiv 2018

  32. [32]

    Y. Tan, C. Chen, W. Zhuang, X. Dong, L. Lyu, G. Long, Is heterogeneity notorious? taming het- erogeneity to handle test-time shift in federated learning, Advances in neural information processing systems 36 (2023) 27167–27180

    2023

  33. [33]

    M. M. Rahimi, H. I. Bhatti, Y. Park, H. Kousar, J. Moon, Evofed: leveraging evolutionary strategies for communication-efficient federated learning, Advances in Neural Information Processing Systems 36 (2023) 62428–62441

    2023

  34. [34]

    H. Zhou, T. Lan, G. P. Venkataramani, W. Ding, Every parameter matters: Ensuring the convergence of federated learning with dynamic heterogeneous models reduction, Advances in Neural Information Processing Systems 36 (2024)

    2024

  35. [35]

    Nguyen, H.-T

    H. Nguyen, H.-T. Nguyen, L. Lovén, S. Pirttikangas, Stake-driven rewards and log-based free rider detection in federated learning, in: 2024 21st Annual International Conference on Privacy, Security and Trust (PST), IEEE, 2024, pp. 1–10

    2024

  36. [36]

    J. Ma, T. Zhou, G. Long, J. Jiang, C. Zhang, Structured federated learning through clustered additive modeling, Advances in Neural Information Processing Systems 36 (2023) 43097–43107

    2023

  37. [37]

    Z. Yang, Y. Zhang, Y. Zheng, X. Tian, H. Peng, T. Liu, B. Han, Fedfed: Feature distillation against data heterogeneity in federated learning, Advances in neural information processing systems 36 (2023) 60397–60428

    2023

  38. [38]

    Zehtabi, D.-J

    S. Zehtabi, D.-J. Han, R. Parasnis, S. Hosseinalipour, C. G. Brinton, Decentralized sporadic fed- erated learning: A unified algorithmic framework with convergence guarantees, in: The Thirteenth International Conference on Learning Representations, 2025

    2025

  39. [39]

    Z. Wang, Z. Wang, L. Lyu, Z. Peng, Z. Yang, C. Wen, R. Yu, C. Wang, X. Fan, Fedsac: Dynamic submodel allocation for collaborative fairness in federated learning, in: Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, 2024, pp. 3299–3310

    2024

  40. [40]

    D. Chen, L. Yao, D. Gao, B. Ding, Y. Li, Efficient personalized federated learning via sparse model- adaptation, in: International conference on machine learning, PMLR, 2023, pp. 5234–5256

    2023

  41. [41]

    Panchal, S

    K. Panchal, S. Choudhary, N. Parikh, L. Zhang, H. Guan, Flow: per-instance personalized federated learning, Advances in Neural Information Processing Systems 36 (2024)

    2024

  42. [42]

    Dorfman, S

    R. Dorfman, S. Vargaftik, Y. Ben-Itzhak, K. Y. Levy, Docofl: Downlink compression for cross-device federated learning, in: International Conference on Machine Learning, PMLR, 2023, pp. 8356–8388. 49

    2023

  43. [43]

    W. Yan, K. Zhang, X. Wang, X. Cao, Problem-parameter-free federated learning, in: The Thirteenth International Conference on Learning Representations, 2025

    2025

  44. [44]

    39879–39902

    R.Ye, M.Xu, J.Wang, C.Xu, S.Chen, Y.Wang, Feddisco: Federatedlearningwithdiscrepancy-aware collaboration, in: International Conference on Machine Learning, PMLR, 2023, pp. 39879–39902

    2023

  45. [45]

    H. Yan, W. Zhang, Q. Chen, X. Li, W. Sun, H. Li, X. Lin, Recess vaccine for federated learning: Proactivedefenseagainstmodelpoisoningattacks, AdvancesinNeuralInformationProcessingSystems 36 (2023) 8702–8713

    2023

  46. [46]

    T. Xia, A. Ghosh, X. Qiu, C. Mascolo, Flea: Addressing data scarcity and label skew in federated learning via privacy-preserving feature augmentation, in: Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, 2024, pp. 3484–3494

    2024

  47. [47]

    W. Bao, H. Wang, J. Wu, J. He, Optimizing the collaboration structure in cross-silo federated learning, in: International Conference on Machine Learning, PMLR, 2023, pp. 1718–1736

    2023

  48. [48]

    X. An, L. Shen, H. Hu, Y. Luo, Federated learning with manifold regularization and normalized update reaggregation, Advances in Neural Information Processing Systems 36 (2023) 55097–55109

    2023

  49. [49]

    Zhang, X

    F. Zhang, X. Liu, S. Lin, G. Wu, X. Zhou, J. Jiang, X. Ji, No one idles: Efficient heterogeneous federated learning with parallel edge and server computation, in: International Conference on Machine Learning, PMLR, 2023, pp. 41399–41413

    2023

  50. [50]

    H. Chen, M. Hao, H. Li, K. Chen, G. Xu, T. Zhang, X. Zhang, Guardhfl: privacy guardian for heterogeneous federated learning, in: International Conference on Machine Learning, PMLR, 2023, pp. 4566–4584

    2023

  51. [51]

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

    2024

  52. [52]

    Z.Liu, Z. Xu, B. Coleman, A.Shrivastava, One-pass distributionsketch for measuring dataheterogene- ity in federated learning, Advances in Neural Information Processing Systems 36 (2023) 15660–15679

    2023

  53. [53]

    3667–3678

    G.Yan, H.Wang, X.Yuan, J.Li, Fedrola: Robustfederatedlearningagainstmodelpoisoningvialayer- based aggregation, in: Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, 2024, pp. 3667–3678

    2024

  54. [54]

    Y. Wu, S. Zhang, W. Yu, Y. Liu, Q. Gu, D. Zhou, H. Chen, W. Cheng, Personalized federated learning under mixture of distributions, in: International Conference on Machine Learning, PMLR, 2023, pp. 37860–37879. 50

    2023

  55. [55]

    Z. Guo, D. Yao, Q. Yang, H. Liu, Hifgl: A hierarchical framework for cross-silo cross-device federated graph learning, in: Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, 2024, pp. 968–979

    2024

  56. [56]

    Kairouz, H

    P. Kairouz, H. B. McMahan, B. Avent, A. Bellet, M. Bennis, A. N. Bhagoji, K. Bonawitz, Z. Charles, G. Cormode, R. Cummings, et al., Advances and open problems in federated learning, Foundations and trends®in machine learning 14 (1–2) (2021) 1–210

    2021

  57. [57]

    Bonawitz, H

    K. Bonawitz, H. Eichner, W. Grieskamp, D. Huba, A. Ingerman, V. Ivanov, C. Kiddon, J. Konečn` y, S. Mazzocchi, B. McMahan, et al., Towards federated learning at scale: System design, Proceedings of machine learning and systems 1 (2019) 374–388

    2019

  58. [58]

    J. Wang, Q. Liu, H. Liang, G. Joshi, H. V. Poor, Tackling the objective inconsistency problem in heterogeneous federated optimization, Advances in neural information processing systems 33 (2020) 7611–7623

    2020

  59. [59]

    Darzidehkalani, M

    E. Darzidehkalani, M. Ghasemi-Rad, P. Van Ooijen, Federated learning in medical imaging: part i: toward multicentral health care ecosystems, Journal of the american college of radiology 19 (8) (2022) 969–974

    2022

  60. [60]

    Linardos, K

    A. Linardos, K. Kushibar, S. Walsh, P. Gkontra, K. Lekadir, Federated learning for multi-center imaging diagnostics: a simulation study in cardiovascular disease, Scientific Reports 12 (1) (2022) 3551

    2022

  61. [61]

    11814–11827

    F.Lai, Y.Dai, S.Singapuram, J.Liu, X.Zhu, H.Madhyastha, M.Chowdhury, Fedscale: Benchmarking model and system performance of federated learning at scale, in: International conference on machine learning, PMLR, 2022, pp. 11814–11827

    2022

  62. [62]

    Ślazyk, P

    F. Ślazyk, P. Jabłecki, A. Lisowska, M. Malawski, S. Płotka, Cxr-fl: deep learning-based chest x- ray image analysis using federated learning, in: International Conference on Computational Science, Springer, 2022, pp. 433–440

    2022

  63. [63]

    C. Yan, L. Meng, L. Li, J. Zhang, Z. Wang, J. Yin, J. Zhang, Y. Sun, B. Zheng, Age-invariant face recognition by multi-feature fusionand decomposition with self-attention, ACM Transactions on Multimedia Computing, Communications, and Applications (TOMM) 18 (1s) (2022) 1–18

    2022

  64. [64]

    Medar, V

    R. Medar, V. S. Rajpurohit, B. Rashmi, Impact of training and testing data splits on accuracy of time series forecasting in machine learning, in: 2017 International Conference on Computing, Communica- tion, Control and Automation (ICCUBEA), IEEE, 2017, pp. 1–6. 51

    2017

  65. [65]

    B. Vrigazova, The proportion for splitting data into training and test set for the bootstrap in clas- sification problems, Business Systems Research: International Journal of the Society for Advancing Innovation and Research in Economy 12 (1) (2021) 228–242

    2021

  66. [66]

    A.Rácz, D.Bajusz, K.Héberger, Effectofdatasetsizeandtrain/testsplitratiosinqsar/qsprmulticlass classification, Molecules 26 (4) (2021) 1111

    2021

  67. [67]

    J. Tan, J. Yang, S. Wu, G. Chen, J. Zhao, A critical look at the current train/test split in machine learning, arXiv preprint arXiv:2106.04525 (2021)

    arXiv 2021

  68. [68]

    K. M. Kahloot, P. Ekler, Algorithmic splitting: A method for dataset preparation, IEEE Access 9 (2021) 125229–125237

    2021

  69. [69]

    J. J. Salazar, L. Garland, J. Ochoa, M. J. Pyrcz, Fair train-test split in machine learning: Mitigating spatialautocorrelationforimprovedpredictionaccuracy, JournalofPetroleumScienceandEngineering 209 (2022) 109885

    2022

  70. [70]

    I. Muraina, Ideal dataset splitting ratios in machine learning algorithms: general concerns for data scientists and data analysts, in: 7th international Mardin Artuklu scientific research conference, 2022, pp. 496–504

    2022

  71. [71]

    V. R. Joseph, A. Vakayil, Split: An optimal method for data splitting, Technometrics 64 (2) (2022) 166–176

    2022

  72. [72]

    Y. Zhao, M. Li, L. Lai, N. Suda, D. Civin, V. Chandra, Federated learning with non-iid data, arXiv preprint arXiv:1806.00582 (2018)

    Pith/arXiv arXiv 2018

  73. [73]

    T. Li, A. K. Sahu, A. Talwalkar, V. Smith, Federated learning: Challenges, methods, and future directions, IEEE signal processing magazine 37 (3) (2020) 50–60

    2020

  74. [74]

    A. Khan, M. Ten Thij, A. Wilbik, Vertical federated learning: A structured literature review, Knowl- edge and Information Systems 67 (4) (2025) 3205–3243

    2025

  75. [75]

    Y. Liu, Y. Kang, T. Zou, Y. Pu, Y. He, X. Ye, Y. Ouyang, Y.-Q. Zhang, Q. Yang, Vertical federated learning: Concepts, advances, and challenges, IEEE Transactions on Knowledge and Data Engineering (2024)

    2024

  76. [76]

    M. Ye, W. Shen, B. Du, E. Snezhko, V. Kovalev, P. C. Yuen, Vertical federated learning for effective- ness, security, applicability: A survey, ACM Computing Surveys 57 (9) (2025) 1–32. 52

    2025

  77. [77]

    T. Feng, D. Bose, T. Zhang, R. Hebbar, A. Ramakrishna, R. Gupta, M. Zhang, S. Avestimehr, S. Narayanan, Fedmultimodal: A benchmark for multimodal federated learning, in: Proceedings of the 29th ACM SIGKDD conference on knowledge discovery and data mining, 2023, pp. 4035–4045

    2023

  78. [78]

    Zhang, L

    T. Zhang, L. Gao, C. He, M. Zhang, B. Krishnamachari, A. S. Avestimehr, Federated learning for the internet of things: Applications, challenges, and opportunities, IEEE Internet of Things Magazine 5 (1) (2022) 24–29

    2022

  79. [79]

    C. Xu, Y. Qu, Y. Xiang, L. Gao, Asynchronous federated learning on heterogeneous devices: A survey, Computer Science Review 50 (2023) 100595

    2023

  80. [80]

    Y. Liu, C. Wang, X. Yuan, Badsampler: Harnessing the power of catastrophic forgetting to poi- son byzantine-robust federated learning, in: Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, 2024, pp. 1944–1955

    2024

Showing first 80 references.