Pith. sign in

REVIEW 4 major objections 5 minor 1 cited by

FedMetaMed: Federated Meta-Learning for Personalized Medication in Distributed Healthcare Systems

T0 review · 4 major / 5 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read FedMetaMed claims to outperform all compared federated learning baselines on skin-lesion classification by aggregating only low-frequency weight components.

desk verdict The method is a plausible new combination, but the paper's own tables don't support the headline results. read the letter →

arxiv 2412.03851 v1 pith:QPRKVRAB submitted 2024-12-05 cs.AI

Jiechao Gao , Yuangang Li This is my paper · ORCID
classification cs.AI
keywords federatedlearningmeta-learningpersonalizedmedicinefrequency-domainaggregationFouriertransformknowledgedistillationmedicalimagingout-of-distributiongeneralization
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

FedMetaMed tries to fix a specific failure mode in federated learning for personalized medicine: when hospitals train on non-identical data, standard parameter averaging makes the global model worse for each site. The paper proposes two coupled mechanisms—Cumulative Fourier Aggregation on the server, which averages only the low-frequency components of client model weights, and Collaborative Transfer Optimization on each client, which transfers global knowledge without overwriting the local personalized model. On two real-world skin-lesion image datasets, the authors report an average F1 of 87% and an average AUC of 90%, outperforming every compared baseline, and an out-of-distribution cohort with 89.51% accuracy and 75.12% F1, against 54.5% F1 for the closest baseline. If these results hold, the recipe offers a privacy-preserving way to build personalized medical models from distributed data without exchanging patient records.

What carries the argument

Cumulative Fourier Aggregation (CFA) is the load-bearing server-side mechanism: it applies a fast Fourier transform to each client's convolutional and fully connected weight tensors, averages only the low-frequency components across clients according to a mask whose radius grows during training, and then inverts the transform, leaving each client's high-frequency components intact. Collaborative Transfer Optimization (CTO) is the client-side mechanism: a second 'client model' receives the server aggregate and is updated through Retrieve (restore with the personalized local model as teacher), Reciprocate (mutual learning between the two models once performance is close), and Refine (the client model teaches the personalized model). The two mechanisms together are intended to accumulate shared low-frequency knowledge while preserving client-specific high-frequency knowledge and local progress.

What would settle it

Run the identical FedMetaMed procedure on the same datasets and splits, but reverse the frequency mask so that high-frequency components are averaged and low-frequency components are kept per client. If the F1 improvement over FedAvg persists, the claimed spectral-semantics mechanism is not what drives the result, and the gains would need a different explanation.

Watch

Extended reading notes

Core claim

The paper's central claim is that the performance drop seen in federated aggregation under data heterogeneity comes from two correctable errors: element-wise averaging in the parameter space, and overwriting local models with the server aggregate. FedMetaMed replaces the first with Cumulative Fourier Aggregation, which transforms each client's weights with the fast Fourier transform, averages only the low-frequency band (a central mask grown from $s_0=0.26$ to $s_1=0.55$ over training), and keeps each client's high-frequency components untouched. It replaces the second with Collaborative Transfer Optimization, a three-step Retrieve–Reciprocate–Refine procedure in which a separate client model receives the server aggregate, is first restored by distillation from the local personalized model, then exchanges knowledge with it, and finally acts as its teacher. On two real-world dermatology datasets, the authors report that this combination gives the best F1 and AUC among all compared baselines, and that an unseen out-of-distribution cohort is classified with 89.51% accuracy and 75.12% F1.

Load-bearing premise

The entire method depends on the idea that the low-frequency parts of a model's weights contain the shareable, general knowledge while the high-frequency parts are client-specific; if that split is wrong, the aggregation could throw away useful knowledge or keep noise, and the reported improvements would not carry over to other settings.

Editorial extensions

If this is right

  • If low-frequency weight components really do carry shared knowledge, CFA could be applied as a drop-in aggregation rule for other federated learning tasks, not just medical imaging.
  • Because CTO never overwrites the local model, clients would not lose their accumulated local knowledge at each communication round, which could improve convergence speed and final accuracy in heterogeneous settings.
  • Reported out-of-distribution F1 of 75.12%, over 20 points higher than the closest baseline, suggests that models trained this way might generalize to hospitals whose patient populations differ from the training cohort.
  • The paper's scaling experiment from 4 to 50 clients indicates the method keeps its advantage as the number of participants grows, which would matter for real hospital networks.

Reading between the lines

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

  • A natural extension the paper does not test: apply CFA to non-convolutional architectures (e.g., attention-based networks), where the frequency decomposition of weight matrices has no established semantic meaning, to see if the low-frequency/high-frequency split still holds.
  • The privacy discussion is qualitative; the paper does not analyze how much information about local data the shared high-frequency parameters can leak, so the frequency split is best read as a heuristic rather than a formal privacy guarantee.
  • The frequency threshold schedule ($s_0=0.26$ to $s_1=0.55$) is fixed by hand; one testable extension is whether the schedule needs to be re-tuned per dataset, architecture, or heterogeneity level for the gains to persist.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper proposes FedMetaMed, a personalized federated meta-learning framework for distributed healthcare data. On the server side, Cumulative Fourier Aggregation (CFA) averages only the low-frequency components of client model parameters while retaining high-frequency components, with the low-frequency threshold increasing over training. On the client side, Collaborative Transfer Optimization (CTO) maintains a deputy model that receives the aggregated server model and transfers knowledge to a personalized local model through Retrieve, Reciprocate, and Refine steps. The authors evaluate FedMetaMed against FedAvg, FedProx, FedBN, SiloBN, IDA, FML, and MetaFed on HAMK and MSK dermoscopic image datasets with four clients, reporting superior F1, AUC, accuracy, and out-of-distribution generalization.

Significance. If the reported gains were reproducible, FedMetaMed would be a useful contribution to personalized federated learning for medical imaging, and the idea of frequency-selective aggregation with a client-side deputy model is conceptually interesting. The paper also includes multiple baselines, ablations, and communication-round curves, which are appropriate experimental components. However, the central empirical claim is not currently verifiable because the summary statistics in Table III do not match the per-client values, the out-of-distribution results are self-contradictory, and no code, data splits, or seeds are provided. As a result, the evidence presented does not support the stated state-of-the-art superiority.

major comments (4)
  1. [Table III] The aggregate F1 and AUC values in Table III cannot be recomputed from the per-client entries. For example, the FedMetaMed F1 values 82.0, 89.7, 83.7, 86.6 average to 85.5, not the reported 87; the MetaFed F1 values 59, 70, 68, 82 average to 69.75, not 76; and the FedBN F1 values 54.6, 72.1, 54.3, 62.7 average to 60.9, not 65.7. The AUC averages are similarly inconsistent: FedBN's 83.0, 96.3, 79, 81 average to 84.8, not 88, and MetaFed's 84, 82, 85, 79 average to 82.5, not 80. Because the paper's headline claim that FedMetaMed outperforms state-of-the-art FL methods rests on these aggregate numbers, the central empirical result is not supported by the manuscript's own data.
  2. [Section IV-D-b, Table II, Table IV] The out-of-distribution generalization results are mutually contradictory. Section IV-D-b states that FedMetaMed achieves an F1 score of 89.50% on the unseen cohort, and Table IV lists an F1 of 89.5, but Table II reports FedMetaMed's F1 on the same out-of-distribution setting as 75.12. The text also claims 'statistically significant improvements' without reporting any statistical tests. These discrepancies make it impossible to determine the true OOD performance and undermine the claim of superior generalization.
  3. [Section IV-D and Table III] The claimed performance margin over FedBN is not consistent with the reported table. Section IV-D says FedMetaMed improves average F1 over FedBN by 9.79%, but Table III reports 87 vs. 65.7, a difference of 21.3 percentage points; even using the corrected FedBN average of 60.9, the difference is 26.1 points. The corresponding AUC improvement is also inconsistent with the reported numbers. This suggests either the comparative figures or the summary statistics are misreported, and the authors should provide corrected tables and text.
  4. [Section IV-B and Section III-A] The main hyperparameters of the proposed method—s0=0.26, s1=0.55, lambda1=0.6, and lambda2=0.8—are fixed on the same two benchmarks used for the final evaluation. No sensitivity analysis, separate validation set, or independent dataset is provided to show that the results are not a product of tuning these threshold parameters. Additionally, Section III-A's core premise that low-frequency components of network parameters determine network capability while high-frequency components carry client-specific information is stated without empirical evidence or theoretical justification. Given that CFA's design depends entirely on this assumption, the paper needs either a spectral analysis of client weights or an ablation across thresholds to support the claim.
minor comments (5)
  1. [Throughout] The method is named 'Cumulative Fourier Agglomeration' in Section III-A and Figure 1 but 'Cumulative Fourier Aggregation' elsewhere; please standardize the terminology.
  2. [Section IV-B] The implementation section reports no random seeds, number of runs, or error bars, and the hardware description includes a typo ('NIVIDIA'). Reporting mean and variance over multiple runs would strengthen the empirical claims.
  3. [Table II] The baseline name is written as 'MetFed' in Table II but 'MetaFed' elsewhere; please make the naming consistent.
  4. [Introduction] The statement that this is 'the first work to apply frequency-domain parameter aggregation in Federated Learning' is not substantiated and is likely inaccurate given existing frequency-domain FL methods; please temper or cite prior work.
  5. [Section IV-D-c] The scaling experiment in Figure 8 reports only four methods and omits several baselines used in the main comparison, such as SiloBN and MetaFed; this limits the strength of the claim about performance with larger client counts.

Circularity Check

0 steps flagged · score 1.0 of 10

No circular derivation found; the empirical claims are benchmark comparisons, and the internal inconsistencies are verification failures rather than circular reductions.

full rationale

FedMetaMed's construction is not circular by definition. CFA (Eq. 3) is a frequency-domain averaging rule over client parameters, and CTO (Eqs. 4-5) is a two-model distillation and mutual-learning procedure; neither equation takes the reported F1, AUC, or accuracy values as an input, and no theorem is invoked that presupposes FedMetaMed's superiority. The low-frequency semantic assumption in Section III-A is an unverified ansatz, but it is not a conclusion derived from its own target claim. The self-citations ([2], [7], [8], [10], [13]) are contextual or motivational and are not load-bearing for the main empirical claim; [13] supports the general observation that element-wise parameter averaging fails under heterogeneity, which is standard federated-learning motivation and independently plausible. The hyperparameters s0=0.26, s1=0.55, lambda1=0.6, lambda2=0.8 are stated as fixed settings in Section IV-B without a described fitting loop on the reported test metrics, so a 'fitted input called prediction' charge cannot be substantiated from the text alone. The paper's headline claim is not fully checkable as printed because Table III's aggregate entries are inconsistent with the per-client values (e.g., MetaFed F1 values 59, 70, 68, 82 average to 69.75, not the reported 76; FedBN AUC values 83.0, 96.3, 79, 81 average to 84.8, not the reported 88), and the out-of-distribution F1 is reported as 75.12 in Table II but as 89.50 in Section IV-D-b and Table IV. Those are internal-evidence failures, not circular reductions: the claims are empirical and externally checkable, but as printed they are not reproducible from the manuscript's own tables. Overall, no derivation step reduces to its own input, so circularity is low; correctness and reproducibility risk should be recorded separately.

Assumptions & free parameters 2 free parameters · 3 assumptions · 1 invented entities

The central method depends on an unproven spectral-semantics assumption about neural network weights, two ad hoc client-side distillation mechanisms, and four hyperparameters tuned on the benchmark. The only new architectural object is an auxiliary per-client model; no physical entities are introduced. These choices are not derived from theory, so the contribution's generality is unestablished.

free parameters (2)
  • s0 and s1 (initial and final low-frequency thresholds) = s0=0.26, s1=0.55
    Set in Section IV-B on HAMK/MSK; the paper gives no sensitivity analysis or criterion for choosing them, so the claimed performance depends on values tuned to these benchmarks.
  • lambda1 and lambda2 (CTO knowledge-transfer thresholds) = lambda1=0.6, lambda2=0.8
    Set in Section IV-B to control when Retrieve, Reciprocate, and Refine switch; no justification or sensitivity study is provided.
assumptions (3)
  • domain assumption Low-frequency components of network parameters encode shared generalizable knowledge and high-frequency components encode client-specific information.
    Section III-A motivates CFA with this claim but provides no spectral analysis of learned weights or ablation over frequency bands.
  • ad hoc to paper A linear cumulative schedule for the low-frequency threshold s stabilizes aggregation and improves convergence.
    The schedule s = s0 + (s1 - s0)/T * t is introduced in Section III-A without proof; the paper claims convergence in the introduction but provides no theorem or proof.
  • ad hoc to paper KL-based mutual distillation between the deputy and personalized models transfers global knowledge without contaminating local personalization.
    Sections III-B to III-E assume the loss terms in Equations 4 and 5 improve the personalized model; no analysis of failure modes or threshold sensitivity is provided.
invented entities (1)
  • Per-client deputy model c (auxiliary model alongside the personalized model q)
    purpose: Receives the server-aggregated model and acts as a bridge in Retrieve-Reciprocate-Refine so global knowledge reaches the personalized model without overwriting local knowledge.
    The deputy model is an architectural addition justified only by the benchmark results in the paper; there is no external falsifiable handle or theoretical analysis establishing that it is necessary.

how reviews work

0 comments
Cite this review

Pith. "Pith review of FedMetaMed: Federated Meta-Learning for Personalized Medication in Distributed Healthcare Systems." pith.science (2026). https://pith.science/paper/QPRKVRAB

@misc{pith2026241203851,
  author       = {Pith},
  title        = {Pith review of: FedMetaMed: Federated Meta-Learning for Personalized Medication in Distributed Healthcare Systems},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/QPRKVRAB}},
  note         = {Machine review of arXiv:2412.03851}
}
read the original abstract

Personalized medication aims to tailor healthcare to individual patient characteristics. However, the heterogeneity of patient data across healthcare systems presents significant challenges to achieving accurate and effective personalized treatments. Ethical concerns further complicate the aggregation of large volumes of data from diverse institutions. Federated Learning (FL) offers a promising decentralized solution by enabling collaborative model training through the exchange of client models rather than raw data, thus preserving privacy. However, existing FL methods often suffer from retrogression during server aggregation, leading to a decline in model performance in real-world medical FL settings. To address data variability in distributed healthcare systems, we introduce Federated Meta-Learning for Personalized Medication (FedMetaMed), which combines federated learning and meta-learning to create models that adapt to diverse patient data across healthcare systems. The FedMetaMed framework aims to produce superior personalized models for individual clients by addressing these limitations. Specifically, we introduce Cumulative Fourier Aggregation (CFA) at the server to improve stability and effectiveness in global knowledge aggregation. CFA achieves this by gradually integrating client models from low to high frequencies. At the client level, we implement a Collaborative Transfer Optimization (CTO) strategy with a three-step process - Retrieve, Reciprocate, and Refine - to enhance the personalized local model through seamless global knowledge transfer. Experiments on real-world medical imaging datasets demonstrate that FedMetaMed outperforms state-of-the-art FL methods, showing superior generalization even on out-of-distribution cohorts.

Figures

Figures reproduced from arXiv: 2412.03851 by the authors.

Figure 1
Figure 1. The diagram illustrates FedMetaMed. The server utilizes Cumulative Fourier Aggregation (CFA) to amalgamate comprehensive knowledge in the [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Dataset distribution percentage across clients. [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 5
Figure 5. Prediction results in terms of F1-Score with number of itera￾tions. 0 50 100 150 200 250 Communication Rounds 0.700 0.725 0.750 0.775 0.800 0.825 0.850 0.875 Average Test Accuracy HAMK Dataset FedAvg FedProx FedBN FML IDA SiloBN FedMetaFed 0 50 100 150 200 250 Communication Rounds 0.700 0.725 0.750 0.775 0.800 0.825 0.850 0.875 0.900 Average Test Accuracy MSK Dataset FedAvg FedProx FedBN FML IDA SiloBN FedMetaFed [… view at source ↗
Figures from the paper (2 more)
Figure 8
Figure 8. Figure 8: Performance of FedMetaMed with State-of-the-Art approaches across [PITH_FULL_IMAGE:figures/full_fig_p007_8.png]
Figure 7
Figure 7. Figure 7: Performance comparison across clients of FedMetaMed method from [PITH_FULL_IMAGE:figures/full_fig_p007_7.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

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

  1. H-FedSN: Personalized Sparse Networks for Efficient and Accurate Hierarchical Federated Learning for IoT Applications

    cs.LG 2024-12 conditional novelty 5.0 of 10

    H-FedSN trains binary masks on frozen weights with shared and private layers, plus Beta-Bernoulli Bayesian aggregation, and reports 58-238x lower communication than HierFAVG with comparable accuracy.

Reference graph

Works this paper leans on

24 extracted references · 17 canonical work pages · cited by 1 Pith paper

  1. [1]

    Dy- namic bank learning for semi-supervised federated image diagnosis with class imbalance,

    M. Jiang, H. Yang, X. Li, Q. Liu, P.-A. Heng, and Q. Dou, “Dy- namic bank learning for semi-supervised federated image diagnosis with class imbalance,” in Medical Image Computing and Computer Assisted Intervention–MICCAI 2022: 25th International Conference, Singapore, September 18–22, 2022, Proceedings, Part III . Springer, 2022, pp. 196–206

    work page 2022

  2. [2]

    Srda: Mobile sensing based fluid overload detection for end stage kidney disease patients using sensor relation dual autoencoder,

    M. Tang, J. Gao, G. Dong, C. Yang, B. Campbell, B. Bowman, J. M. Zoellner, E. Abdel-Rahman, and M. Boukhechba, “Srda: Mobile sensing based fluid overload detection for end stage kidney disease patients using sensor relation dual autoencoder,” in Conference on Health, Inference, and Learning. PMLR, 2023, pp. 133–146

    2023

  3. [3]

    Dfml: Dynamic federated meta-learning for rare disease prediction,

    B. Chen, T. Chen, X. Zeng, W. Zhang, Q. Lu, Z. Hou, J. Zhou, and S. Helal, “Dfml: Dynamic federated meta-learning for rare disease prediction,” IEEE/ACM Transactions on Computational Biology and Bioinformatics, 2023

    work page 2023

  4. [4]

    Artificial intelligence-aided digital twin design: A systematic review,

    N. Hao, Y . Li, K. Liu, S. Liu, Y . Lu, B. Xu, C. Li, J. Chen, L. Yue, T. Fu, X. Hu, X. Wang, and Y . Zhao, “Artificial intelligence-aided digital twin design: A systematic review,” Preprints, August 2024. [Online]. Available: https://doi.org/10.20944/preprints202408.2063.v1

    work page arXiv 2024

  5. [5]

    Federated learning systems for healthcare: perspective and recent progress,

    Y . Kumar and R. Singla, “Federated learning systems for healthcare: perspective and recent progress,” Federated Learning Systems: Towards Next-Generation AI, pp. 141–156, 2021

    work page 2021

  6. [6]

    Fedbcgd: Communication-efficient accelerated block coordinate gradient descent for federated learning,

    J. Liu, F. Shang, Y . Liu, H. Liu, Y . Li, and Y . Gong, “Fedbcgd: Communication-efficient accelerated block coordinate gradient descent for federated learning,” in Proceedings of the 32nd ACM International Conference on Multimedia , 2024

    work page 2024

  7. [7]

    Pfed-ldp: A personalized federated local differential privacy framework for iot sensing data,

    J. Gao, M. Tang, T. Wang, and B. Campbell, “Pfed-ldp: A personalized federated local differential privacy framework for iot sensing data,” in Proceedings of the 20th ACM Conference on Embedded Networked Sensor Systems, 2022, pp. 835–836

    work page 2022

  8. [8]

    Residential energy management system using personalized federated deep reinforcement learning,

    J. Gao, W. Wang, and B. Campbell, “Residential energy management system using personalized federated deep reinforcement learning,” in 2022 21st ACM/IEEE International Conference on Information Process- ing in Sensor Networks (IPSN) . IEEE, 2022, pp. 541–542

    work page 2022

Show all 24 references
  1. [9]

    Difficulty- aware meta-learning for rare disease diagnosis,

    X. Li, L. Yu, Y . Jin, C.-W. Fu, L. Xing, and P.-A. Heng, “Difficulty- aware meta-learning for rare disease diagnosis,” in Medical Image Computing and Computer Assisted Intervention–MICCAI 2020: 23rd International Conference, Lima, Peru, October 4–8, 2020, Proceedings, Part I ...

    2020

  2. [10]

    Pfdrl: Personalized federated deep reinforcement learning for residen- tial energy management,

    J. Gao, W. Wang, F. Nikseresht, V . Govinda Rajan, and B. Campbell, “Pfdrl: Personalized federated deep reinforcement learning for residen- tial energy management,” in Proceedings of the 52nd International Conference on Parallel Processing, 2023, pp. 402–411

    2023

  3. [11]

    Pmfl: Partial meta-federated learning for heterogeneous tasks and its applications on real-world medical records,

    T. Zhang, S. Zhang, Z. Chen, Y . Bengio, and D. Liu, “Pmfl: Partial meta-federated learning for heterogeneous tasks and its applications on real-world medical records,” in 2022 IEEE International Conference on Big Data (Big Data) . IEEE, 2022, pp. 4453–4462

    2022

  4. [12]

    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, p. 106775, 2021

    2021

  5. [13]

    Fed-ldr: Federated local data-infused graph creation with node-centric model refinement,

    J. Gao, Y . Li, and S. F. Ahmed, “Fed-ldr: Federated local data-infused graph creation with node-centric model refinement,” 2024

    2024

  6. [14]

    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,” arXiv preprint arXiv:1907.02189 , 2019

    1907 arXiv

  7. [15]

    Federated optimization in heterogeneous networks,

    T. Li, A. K. Sahu, M. Zaheer, M. Sanjabi, A. Talwalkar, and V . Smith, “Federated optimization in heterogeneous networks,” 2020

    2020

  8. [16]

    Federated learning of predictive models from federated electronic health records,

    T. S. Brisimi, R. Chen, T. Mela, A. Olshevsky, I. C. Paschalidis, and W. Shi, “Federated learning of predictive models from federated electronic health records,” International journal of medical informatics , vol. 112, pp. 59–67, 2018

    2018

  9. [17]

    Fedbn: Feder- ated learning on non-iid features via local batch normalization,

    X. Li, M. Jiang, X. Zhang, M. Kamp, and Q. Dou, “Fedbn: Feder- ated learning on non-iid features via local batch normalization,” arXiv preprint arXiv:2102.07623, 2021

    2021 arXiv

  10. [18]

    Siloed federated learning for multi-centric histopathology datasets,

    M. Andreux, J. O. du Terrail, C. Beguier, and E. W. Tramel, “Siloed federated learning for multi-centric histopathology datasets,” in Domain Adaptation and Representation Transfer, and Distributed and Collabo- rative Learning: Second MICCAI Workshop, DART 2020 . Springer, 2020...

    2020

  11. [19]

    Fedhealth: A federated transfer learning framework for wearable healthcare,

    Y . Chen, X. Qin, J. Wang, C. Yu, and W. Gao, “Fedhealth: A federated transfer learning framework for wearable healthcare,” IEEE Intelligent Systems, vol. 35, no. 4, pp. 83–93, 2020

    2020

  12. [20]

    On bridging generic and person- alized federated learning for image classification,

    H.-Y . Chen and W.-L. Chao, “On bridging generic and person- alized federated learning for image classification,” arXiv preprint arXiv:2107.00778, 2021

    2021 arXiv

  13. [21]

    Federated reinforcement learning-supported ids for iot-steered healthcare systems,

    S. Otoum, N. Guizani, and H. Mouftah, “Federated reinforcement learning-supported ids for iot-steered healthcare systems,” in ICC 2021- IEEE International Conference on Communications . IEEE, 2021, pp. 1–6

    2021

  14. [22]

    Protecting personal healthcare record using blockchain & federated learning technologies,

    S. Aich, N. K. Sinai, S. Kumar, M. Ali, Y . R. Choi, M.-I. Joo, and H.-C. Kim, “Protecting personal healthcare record using blockchain & federated learning technologies,” in 2022 24th International Conference on Advanced Communication Technology (ICACT) . IEEE, 2022, pp. 109–112

    2022

  15. [23]

    Metafed: Feder- ated learning among federations with cyclic knowledge distillation for personalized healthcare,

    Y . Chen, W. Lu, X. Qin, J. Wang, and X. Xie, “Metafed: Feder- ated learning among federations with cyclic knowledge distillation for personalized healthcare,” IEEE Transactions on Neural Networks and Learning Systems, 2023

    2023

  16. [24]

    Federated meta- learning enhanced acoustic radio cooperative framework for ocean of things,

    H. Zhao, F. Ji, Q. Li, Q. Guan, S. Wang, and M. Wen, “Federated meta- learning enhanced acoustic radio cooperative framework for ocean of things,” IEEE Journal of Selected Topics in Signal Processing , vol. 16, no. 3, pp. 474–486, 2022

    2022

Pith tools

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