Pith. sign in

REVIEW 4 major objections 6 minor 50 references

CADRE: Customizable Assurance of Data Readiness in Privacy-Preserving Federated Learning

T0 review · 4 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read CADRE lets FL administrators define custom data-readiness metrics, rules, and remedies that run locally on each client and produce aggregated privacy-preserving DR reports, improving downstream model performance.

desk verdict A genuinely useful customizable data-readiness framework for PPFL, but the privacy-preservation claim needs real accounting before it can be taken at face value. read the letter →

arxiv 2505.23849 v2 pith:P6VDDCZV submitted 2025-05-28 cs.CR cs.AIcs.LG

Kaveen Hiniduma , Zilinghan Li , Aditya Sinha , Ravi Madduri , Suren Byna This is my paper
classification cs.CRcs.AIcs.LG
keywords datareadinessfederatedlearningprivacy-preservingqualityassessmentcustommetricsremediesPPFLDRreports
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper claims that the biggest obstacle to using federated learning in practice—administrators cannot see or clean the data—can be removed without breaking privacy. The proposed framework, CADRE, lets each FL administrator define their own data-readiness metrics, rules, and remedies; each client runs those functions on its own data, and only aggregated statistics and plots go back to the server as a DR report. The paper reports that after applying CADRE, nearly every polluted dataset met the admin-defined standards, and on a brain-MRI segmentation task the noise-handling module improved and stabilized the average Dice score across clients. The point of the work is that data quality assurance can be made customizable, local, and privacy-preserving, and that doing so pays off in downstream model performance.

What carries the argument

The central object is the CADRE module, a template with three extensible functions: metrics (numeric evaluations of the client's data, e.g., mean intensity, imbalance degree, k-anonymity level), rules (thresholds that decide whether the data is 'AI-ready'), and remedies (actions applied locally, such as removal of noisy indices, SMOTE oversampling, or suppression of low-anonymity records). The machinery does its work by running each client through the evaluate–rule–remedy loop before training starts, then compiling standard metric plots, custom metric values, and a combined PCA plot from all clients into an aggregated HTML DR report for the administrator. Because execution is local and the report contains only aggregated statistics, the framework claims to preserve the privacy guarantees of the underlying PPFL system.

What would settle it

Run CADRE on a dataset with known sensitive attributes, collect the aggregated per-client class distributions and PCA projections from the DR reports, and attempt a membership-inference or attribute-inference attack; if an adversary can reconstruct a client's sensitive values (e.g., income bracket from class distribution + PCA) with accuracy above the DP baseline, the central privacy claim fails. A simpler check: apply k-anonymity to the raw data, then verify whether the report's aggregates themselves satisfy k-anonymity.

Watch

Extended reading notes

Core claim

The central claim is that a customizable, locally executed data-readiness layer can sit inside an existing privacy-preserving federated learning stack and improve both compliance and model quality. CADRE's contribution is the separation of concerns: administrators define the evaluation logic (metrics), the pass/fail conditions (rules), and the fixes (remedies); clients execute all of it on raw data that never leaves the device; and the server receives only aggregated metric values, class distributions, and PCA projections, which the paper asserts preserve privacy. The evaluation shows that for seven DR issues—noise, class imbalance, duplicates, memory bloat, fairness gaps, outliers, and weak k-anonymity—the custom modules brought nearly all client datasets back within the defined thresholds, and the single exception (a client with only one ethnic group that could not be rebalanced) was visible in the DR report. The paper's strongest downstream result is on the Flamby IXI Tiny dataset, where applying the noise-handling CADRE module before training yields consistently higher and more stable Dice scores over ten rounds compared to training without it.

Load-bearing premise

The framework's privacy claim rests on the assumption that the aggregated statistics sent to the server—means, class distributions, and PCA plots—cannot be combined to reveal individual client records, yet no differential-privacy or attack analysis is provided to back that up.

Editorial extensions

If this is right

  • Federated learning deployments in healthcare can set per-hospital image-quality standards (resolution, noise threshold, format) and have those standards enforced automatically before training begins.
  • Data stewards can iterate: if a DR report shows a client was flagged but the remedy failed (e.g., a single-ethnic-group client), they can adjust the rule or exclude the client before spending compute.
  • The same module structure can be reused across modalities—2D images, tabular data, 3D volumes—by swapping in task-specific metrics and remedies, reducing the cost of data preparation.
  • Resource-constrained edge FL clients benefit because memory-footprint rules can downcast data types or drop duplicates before they are sent to training.

Reading between the lines

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

  • If aggregated DR statistics (e.g., per-client class distributions and PCA projections) are shown to leak information under repeated queries, CADRE's privacy claim would need a differential-privacy layer; the paper does not provide that accounting.
  • One could extend CADRE to run continuously during training, not just pre-training, flagging drift in a client's data distribution from one round to the next using the same metric/rule/remedy pattern.
  • The performance lift shown on IXI Tiny suggests a testable design rule: for noisy modalities, a pre-training noise-filtering remedy may dominate in-sample data augmentation, because it removes rather than re-weights corrupted labels.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 6 minor

Summary. The paper introduces CADRE, a framework for customizable data-readiness (DR) assurance in privacy-preserving federated learning (PPFL). It allows FL administrators to define custom metrics, rules, and remedies, which clients execute locally, and produces aggregated DR reports for administrator review. The authors integrate CADRE into the APPFL framework and evaluate it on six datasets and seven DR modules, covering noise, class imbalance, duplicates, memory, bias, outliers, and k-anonymity. The central claims are that CADRE preserves privacy by exchanging only aggregated statistics, and that it improves downstream FL model performance, demonstrated by a Dice-score experiment on the IXI Tiny dataset.

Significance. If the privacy claim and the performance improvement were rigorously supported, CADRE would be a useful contribution: it addresses a real gap in PPFL practice, namely the lack of tooling for pre-training data readiness in a decentralized setting. The paper's strengths include the concrete implementation and integration with APPFL, the breadth of datasets and modules (six datasets, seven modules), and the availability of code and documentation. The framework's extensibility and the local execution of metrics/remedies are demonstrated convincingly. However, the load-bearing privacy guarantee is asserted rather than established, and the downstream performance claim rests on a single experiment with no statistical validation or baseline comparison. These gaps currently prevent the paper from supporting its central value proposition.

major comments (4)
  1. [§III-C, contribution bullet 2, Fig. 2] The privacy-preservation claim is unsupported. The DR report contains per-client class distributions, statistical moments, k-anonymity levels, and a combined PCA plot generated from 'a sample of the data from the clients.' The paper provides no differential-privacy accounting, noise calibration, or attack analysis for these artifacts. In a standard PPFL threat model, exact per-client statistics (e.g., class counts revealing disease prevalence at a hospital) and PCA projections can leak information beyond what is revealed by model updates. Because the abstract's 'while preserving privacy' is a central promise, the authors must either provide formal privacy guarantees for the DR report, restrict the report to privacy-preserving aggregates (e.g., with secure aggregation or DP noise), or clearly delimit the threat model under which the report is considered privacy-preserving.
  2. [§III-C, Fig. 2, contribution bullet 2] The description of the combined PCA plot appears to contradict the 'no raw data' claim. The text states the PCA plot is based on 'a sample of the data from the clients,' while the contribution bullet claims the report includes 'only aggregated metric evaluations without exposing any raw data.' If raw data points (or near-raw transformed points) are sent to the server to create the PCA plot, the central privacy claim collapses; if the plot is generated from aggregates or via a privacy-preserving protocol, the manuscript must state this explicitly and specify the mechanism. As written, the contradiction is load-bearing and unresolved.
  3. [§V-D, Fig. 8] The downstream performance claim is based on a single experiment (IXI Tiny, noise management module) over 10 rounds with no statistical test and no comparison to alternative quality-control baselines (e.g., simple outlier clipping, median filtering, or an existing method like FedDQA). The statement 'applying CADRE leads to consistently improved and more stable Dice scores' is not supported without multiple runs, confidence intervals, or a significance test. This experiment is the only independent evidence that remedies improve model performance; the conclusion should be tempered accordingly or the evaluation strengthened.
  4. [Table I, Fig. 7] The before/after metric comparisons in Figure 7 are largely tautological. Each remedy (e.g., removing noisy indices, SMOTE oversampling, duplicate removal) is defined to reduce the very metric it targets, so the observation that post-remedy metrics move toward the threshold mostly confirms that the code performs its intended operation. The independent evidence of real-world effectiveness must come from downstream task performance, which is currently limited to the IXI Tiny experiment. The paper should either present downstream results for more modules/datasets or explicitly frame Figure 7 as a functional correctness check rather than evidence of effectiveness.
minor comments (6)
  1. [Figure 2] The caption and text refer to panels (a)-(d), but the figure layout is not clearly described; consider labeling the subfigures directly in the image and describing each in the text.
  2. [Table I, Module 1] The rule threshold (e.g., mean >0.37 for MNIST) is presented as an example, but the paper does not explain how such thresholds should be chosen in practice; a brief guidance or a reference to a heuristic would improve reproducibility.
  3. [§V-C, Table II] Adding Gaussian noise with std.dev. = 2 to 90% of the data may create extreme pollution for image data (e.g., MNIST and CIFAR-10 values are typically in [0,1]); clarify whether the noise is added in a normalized space and whether the chosen magnitude is realistic.
  4. [§I] The term 'administrators' is introduced in quotes and defined informally; a more precise definition at first use would help avoid ambiguity with server-side FL operators.
  5. [Abstract] The phrase 'utilizes valuable resources' is vague and could be misinterpreted; consider rephrasing to 'conserves computational and organizational resources.'
  6. [References] Reference [9] is noted as 'just Accepted'; update the citation with the final publication details if available at the time of revision.

Circularity Check

2 steps flagged · score 5.0 of 10

The evaluation of CADRE's remedies is partly tautological: each remedy moves the same metric its rule monitors, so the Figure 7 before/after plots largely confirm the code's own definitions. The privacy guarantee is also asserted by equating 'no raw data' with 'privacy preservation' without leakage analysis. Independent content remains in the IXI Tiny Dice experiment.

  1. self definitional [Section V-D (Results) and Table I (Custom CADRE Modules)]
    "Figure 7 illustrates the metric values before and after applying the remedies of custom CADRE modules, with threshold values indicating the rules set for each experiment. As shown in the figure, almost all post-remedy data points fall within the expected range. Module 2: Applied when imbalance degree>0. SMOTE was used to oversample the minority class. Module 7: Applied when anonymity level≤1. Data records with low anonymity levels were suppressed to ensure the desired level of anonymity."

    The before/after comparisons in Figure 7 are not independent predictions: each remedy is triggered by and evaluated on the same quantity. Module 2 uses class-imbalance degree as the rule and SMOTE as the remedy, so the post-remedy plot shows the imbalance dropping because the minority class was oversampled; Module 7 uses k-anonymity as the rule and record suppression as the remedy, so the post-remedy level rises because suppression is designed to make records less distinguishable. The one red-box failure is an edge case, not a counterexample to the tautology. The separate Dice-score experiment (Figure 8) provides the only genuinely non-constructed evidence of downstream benefit.

  2. self definitional [Section I, contribution bullet 2; Section III-C (DR Reporting Module)]
    "We generate comprehensive DR reports in CADRE that evaluate the metrics defined by FL administrators. This ensures privacy preservation by only including aggregated metric evaluations without exposing any raw data."

    The sentence equates privacy preservation with the absence of raw data and the use of aggregates. That stipulated equivalence is the entire argument: the DR report 'ensures privacy preservation' because it sends only aggregates. No DP accounting, noise calibration, or membership/attribute-inference analysis is provided for the per-client statistics, k-anonymity levels, or combined PCA plot. Moreover, Section III-C says the combined plot is a PCA visualization of a sample of the data from clients, which sits in tension with the no-raw-data premise. The privacy conclusion is therefore the definitional premise restated, not a derived guarantee.

full rationale

The paper is a framework paper rather than a mathematical derivation, so most of it is not a prediction chain. The clearest circularity is the validation logic: Figure 7's before/after metrics are generated by modules whose rules and remedies operate on the same metric, so those plots mainly verify that the code does what it was told to do. That is genuine but partial self-referentiality. The second issue is the privacy claim, which is load-bearing for the PPFL contribution and is asserted as an equivalence between 'aggregated/no raw data' and 'privacy preservation' without leakage analysis; I treat this as a self-definitional step rather than a mere correctness gap because the paper gives no independent argument connecting the released aggregates to the PPFL threat model. I do not raise the score to 8 because the IXI Tiny Dice experiment (Figure 8) is externally benchmarked and not constructed by a rule/remedy pair, and the extensibility/customizability contribution does not depend on a self-citation. The self-citations to AIDRIN [10] and the survey [9] are not load-bearing: they supply standard statistical metrics and a taxonomy, not a uniqueness theorem that forces the framework's design.

Assumptions & free parameters 1 free parameters · 2 assumptions · 0 invented entities

The central claim rests on two unproved domain assumptions: that aggregated statistics do not leak private information, and that local remediation preserves PPFL privacy semantics. Additionally, administrator-chosen rule thresholds determine the demonstrated success. No new physical or mathematical entities are introduced.

free parameters (1)
  • Rule thresholds in CADRE modules = e.g., mean > 0.37 for MNIST noise; imbalance degree > 0; duplicate proportion > 0; anonymity level <= 1
    Administrator-defined thresholds in Table I. The demonstrated before/after success and the claimed performance improvement depend on these hand-chosen values, although the framework itself treats them as configuration.
assumptions (2)
  • domain assumption Aggregated per-client statistics (means, class distributions, PCA) are privacy-preserving.
    Used to claim CADRE preserves privacy (Section III-C and contribution bullet). No differential-privacy accounting or attack analysis is provided.
  • domain assumption Local execution of metrics and remedies is compatible with PPFL privacy guarantees.
    Remedies run on client data before training (Section IV); the framework assumes this does not alter the privacy properties of the downstream federated learning pipeline.

how reviews work

0 comments
Cite this review

Pith. "Pith review of CADRE: Customizable Assurance of Data Readiness in Privacy-Preserving Federated Learning." pith.science (2026). https://pith.science/paper/P6VDDCZV

@misc{pith2026250523849,
  author       = {Pith},
  title        = {Pith review of: CADRE: Customizable Assurance of Data Readiness in Privacy-Preserving Federated Learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/P6VDDCZV}},
  note         = {Machine review of arXiv:2505.23849}
}
read the original abstract

Privacy-Preserving Federated Learning (PPFL) is a decentralized machine learning approach where multiple clients train a model collaboratively. PPFL preserves the privacy and security of a client's data without exchanging it. However, ensuring that data at each client is of high quality and ready for federated learning (FL) is a challenge due to restricted data access. In this paper, we introduce CADRE (Customizable Assurance of Data Readiness) for federated learning (FL), a novel framework that allows users to define custom data readiness (DR) metrics, rules, and remedies tailored to specific FL tasks. CADRE generates comprehensive DR reports based on the user-defined metrics, rules, and remedies to ensure datasets are prepared for FL while preserving privacy. We demonstrate a practical application of CADRE by integrating it into an existing PPFL framework. We conducted experiments across six datasets and addressed seven different DR issues. The results illustrate the versatility and effectiveness of CADRE in ensuring DR across various dimensions, including data quality, privacy, and fairness. This approach enhances the performance and reliability of FL models as well as utilizes valuable resources.

Figures

Figures reproduced from arXiv: 2505.23849 by the authors.

Figure 1
Figure 1. An overview of CADRE framework for FL tasks. [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. The figure illustrates an example DR report from an FL [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. An iterative data evaluation and remediation process [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (4 more)
Figure 5
Figure 5. Figure 5: YAML configuration for customizing a CADRE mod [PITH_FULL_IMAGE:figures/full_fig_p005_5.png]
Figure 6
Figure 6. Figure 6: Example DR reports generated before (left) and after (right) applying CADRE module 1 show an improvement in the [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]
Figure 7
Figure 7. Figure 7: Evaluation of custom metrics for each CADRE module, before and after remedy application. Threshold lines indicate [PITH_FULL_IMAGE:figures/full_fig_p007_7.png]
Figure 8
Figure 8. Figure 8: The figure presents a comparative analysis of FL [PITH_FULL_IMAGE:figures/full_fig_p008_8.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

50 extracted references · 35 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,”Artificial intelligence and statistics, pp. 1273–1282, 2017

    work page 2017

  2. [2]

    Federated learning: Challenges, methods, and future directions,

    T. Li, A. K. Sahu, A. Talwalkar, and V . Smith, “Federated learning: Challenges, methods, and future directions,”IEEE Signal Processing Magazine, vol. 37, no. 3, pp. 50–60, 2020

    2020

  3. [3]

    Data pipeline challenges in privacy-preserving federated learn- ing,

    X. Huang, Y . Dong, and S. Pentyala, “Data pipeline challenges in privacy-preserving federated learn- ing,” https://www.nist.gov/blogs/cybersecurity-insights/ data-pipeline-challenges-privacy-preserving-federated-learning, February 2024, nIST Cybersecurity Insights Blog Post. Part of a series on privacy-preserving federated learning in collaboration with ...

    work page 2024

  4. [4]

    The impact of data quality on federated versus centralized learning,

    G. Nilsson, “The impact of data quality on federated versus centralized learning,” Master of Science in Engineering: AI and Machine Learning, Blekinge Institute of Technology, 371 79 Karlskrona, Sweden, July 2024

    work page 2024

  5. [5]

    The role of the data quality on model efficiency: An exploratory study on centralised and federated learning,

    G. Nilsson, M. Boldt, and S. Alawadi, “The role of the data quality on model efficiency: An exploratory study on centralised and federated learning,” in2024 9th International Conference on Fog and Mobile Edge Computing (FMEC), 2024, pp. 253–260

    work page 2024

  6. [6]

    Enhancing data quality in federated fine-tuning of foundation models,

    W. Zhao, Y . Du, N. D. Lane, S. Chen, and Y . Wang, “Enhancing data quality in federated fine-tuning of foundation models,”arXiv preprint arXiv:2403.04529, Mar 2024

    arXiv 2024

  7. [7]

    EIFFeL: Ensuring Integrity for Federated Learning

    A. R. Chowdhury, C. Guo, S. Jha, and L. van der Maaten, “Eiffel: Ensuring integrity for federated learning,” 2022. [Online]. Available: https://arxiv.org/abs/2112.12727

    work page Pith review arXiv 2022

  8. [8]

    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,” 2022. [Online]. Available: https://arxiv.org/abs/2012.13995

    arXiv 2022

Show all 50 references
  1. [9]

    Data readiness for ai: A 360-degree survey,

    K. Hiniduma, S. Byna, and J. L. Bez, “Data readiness for ai: A 360-degree survey,”ACM Comput. Surv., Mar. 2025, just Accepted. [Online]. Available: https://doi.org/10.1145/3722214

    2025 doi

  2. [10]

    Ai data readiness inspector (aidrin) for quantitative assessment of data readiness for ai,

    K. Hiniduma, S. Byna, J. L. Bez, and R. Madduri, “Ai data readiness inspector (aidrin) for quantitative assessment of data readiness for ai,” inProceedings of the 36th International Conference on Scientific and Statistical Database Management, ser. SSDBM ’24. New York, NY , US...

    2024

  3. [11]

    Enabling end-to- end secure federated learning in biomedical research on heterogeneous computing environments with appflx,

    T.-H. Hoang, J. Fuhrman, M. Klarqvist, M. Li, P. Chaturvedi, Z. Li, K. Kim, M. Ryu, R. Chard, E. A. Huertaet al., “Enabling end-to- end secure federated learning in biomedical research on heterogeneous computing environments with appflx,”Computational and Structural Biotechnol...

    2025

  4. [12]

    Advances in appfl: A comprehensive and extensible federated learning framework,

    Z. Li, S. He, Z. Yang, M. Ryu, K. Kim, and R. Madduri, “Advances in appfl: A comprehensive and extensible federated learning framework,” arXiv preprint arXiv:2409.11585, 2024

    2024 arXiv

  5. [13]

    Appfl: open-source software framework for privacy-preserving federated learning,

    M. Ryu, Y . Kim, K. Kim, and R. K. Madduri, “Appfl: open-source software framework for privacy-preserving federated learning,” in2022 IEEE International Parallel and Distributed Processing Symposium Workshops (IPDPSW). IEEE, 2022, pp. 1074–1083

    2022

  6. [14]

    Dqlearn: A toolkit for structured data quality learning,

    S. Shrivastavaet al., “Dqlearn: A toolkit for structured data quality learning,” inProceedings of the IEEE International Conference on Big Data (Big Data), 2020, pp. 1644–1653

    2020

  7. [15]

    Data quality toolkit: Automatic assessment of data quality and remediation for machine learning datasets,

    N. Gupta, H. Patelet al., “Data quality toolkit: Automatic assessment of data quality and remediation for machine learning datasets,”arXiv preprint arXiv:2108.05935, 2021

    2021 arXiv

  8. [16]

    Data readiness report,

    S. Afzal, C. Rajmohan, M. Kesarwani, S. Mehta, and H. Patel, “Data readiness report,” inProceedings of the IEEE International Conference on Smart Data Services (SMDS), 2020, pp. 42–51

    2020

  9. [17]

    Automating large-scale data quality verification,

    S. Schelter, D. Lange, P. Schmidt, M. Celikel, F. Biessmann, and A. Grafberger, “Automating large-scale data quality verification,”Pro- ceedings of the VLDB Endowment, vol. 11, no. 12, pp. 1781–1794, August 2018

    2018

  10. [18]

    Holoclean: holistic data repairs with probabilistic inference,

    T. Rekatsinas, X. Chu, I. F. Ilyas, and C. R ´e, “Holoclean: holistic data repairs with probabilistic inference,”Proc. VLDB Endow., vol. 10, no. 11, p. 1190–1201, Aug. 2017. [Online]. Available: https://doi.org/10.14778/3137628.3137631

    2017

  11. [19]

    Data quality sla rule compliance and remediation,

    IBM, “Data quality sla rule compliance and remediation,” https://dataplatform.cloud.ibm.com/docs/content/wsj/quality/ dq-sla-compliance.html?context=cpdaas&audience=wdp, 2015, accessed: 2025-04-30

    2015

  12. [20]

    Tackling noisy clients in federated learning with end-to-end label correction,

    X. Jiang, S. Sun, J. Li, J. Xue, R. Li, Z. Wu, G. Xu, Y . Wang, and M. Liu, “Tackling noisy clients in federated learning with end-to-end label correction,” inProceedings of the 33rd ACM International Conference on Information and Knowledge Management, ser. CIKM ’24. ACM, Oct....

    2024

  13. [21]

    Feddqa: A novel regularization-based deep learning method for data quality assessment in federated learning,

    Z. Zhang, G. Chen, Y . Xu, L. Huang, C. Zhang, and S. Xiao, “Feddqa: A novel regularization-based deep learning method for data quality assessment in federated learning,”Decision Support Systems, vol. 180, p. 114183, 2024. [Online]. Available: https: //doi.org/10.1016/j.dss.20...

    2024

  14. [22]

    Lia: Privacy-preserving data quality evaluation in federated learning using a lazy influence approximation,

    L. Rokvic, P. Danassis, S. P. Karimireddy, and B. Faltings, “Lia: Privacy-preserving data quality evaluation in federated learning using a lazy influence approximation,” 2024. [Online]. Available: https://arxiv.org/abs/2205.11518

    2024 arXiv

  15. [23]

    Data quality control in federated instruction-tuning of large language models,

    Y . Du, R. Ye, F. Yuchi, W. Zhao, J. Qu, Y . Wang, and S. Chen, “Data quality control in federated instruction-tuning of large language models,” 2025. [Online]. Available: https://arxiv.org/abs/2410.11540

    2025 arXiv

  16. [24]

    Industry 5.0,

    European Commission, “Industry 5.0,” 2021, accessed: 2025-04-

    2021

  17. [25]

    Available: https://research-and-innovation.ec.europa.eu/ research-area/industrial-research-and-innovation/industry-50 en

    [Online]. Available: https://research-and-innovation.ec.europa.eu/ research-area/industrial-research-and-innovation/industry-50 en

  18. [26]

    On lines and planes of closest fit to systems of points in space,

    K. Pearson, “On lines and planes of closest fit to systems of points in space,”The London, Edinburgh, and Dublin Philosophical Magazine and Journal of Science, vol. 2, no. 11, pp. 559–572, 1901

    1901

  19. [27]

    A survey on class imbalance in federated learning,

    J. Zhang, C. Li, J. Qi, and J. He, “A survey on class imbalance in federated learning,” 2023. [Online]. Available: https://arxiv.org/abs/ 2303.11673

    2023 arXiv

  20. [28]

    Kohavi, “Adult,” UCI Machine Learning Repository, 1996, dOI: 10.24432/C5GP7S

    R. Kohavi, “Adult,” UCI Machine Learning Repository, 1996, dOI: 10.24432/C5GP7S. [Online]. Available: https://doi.org/10.24432/ C5GP7S

    1996 doi

  21. [29]

    Measuring the effects of non- identical data distribution for federated visual classification,

    T.-M. H. Hsu, H. Qi, and M. Brown, “Measuring the effects of non- identical data distribution for federated visual classification,”arXiv preprint arXiv:1909.06335, 2019

    1909 arXiv

  22. [30]

    Federated learning with buffered asynchronous aggregation,

    J. Nguyen, K. Malik, H. Zhan, A. Yousefpour, M. Rabbat, M. Malek, and D. Huba, “Federated learning with buffered asynchronous aggregation,” inInternational Conference on Artificial Intelligence and Statistics. PMLR, 2022, pp. 3581–3607

    2022

  23. [31]

    FedCompass: efficient cross-silo federated learning on heterogeneous client devices using a computing power aware scheduler,

    Z. Li, P. Chaturvedi, S. He, H. Chen, G. Singh, V . Kindratenko, E. A. Huerta, K. Kim, and R. Madduri, “FedCompass: efficient cross-silo federated learning on heterogeneous client devices using a computing power aware scheduler,”arXiv preprint arXiv:2309.14675, 2023

    2023 arXiv

  24. [32]

    Differential privacy,

    C. Dwork, “Differential privacy,” inInternational colloquium on au- tomata, languages, and programming. Springer, 2006, pp. 1–12

    2006

  25. [33]

    Data readiness assurance framework in appfl,

    APPFL Contributors, “Data readiness assurance framework in appfl,” https://appfl.ai/en/latest/tutorials/examples dr integration.html, 2025

    2025

  26. [34]

    An experimental study of class imbalance in federated learning,

    C. Xiao and S. Wang, “An experimental study of class imbalance in federated learning,” in2021 IEEE Symposium Series on Computational Intelligence (SSCI). IEEE, 2021

    2021

  27. [35]

    Smote: Synthetic minority over-sampling technique,

    N. V . Chawla, K. W. Bowyer, L. O. Hall, and W. P. Kegelmeyer, “Smote: Synthetic minority over-sampling technique,”Journal of Artificial Intel- ligence Research, vol. 16, pp. 321–357, 2002

    2002

  28. [36]

    Fairness through awareness,

    C. Dwork, M. Hardt, T. Pitassi, O. Reingold, and R. Zemel, “Fairness through awareness,”Proceedings of the 3rd Innovations in Theoretical Computer Science Conference, pp. 214–226, 2012

    2012

  29. [37]

    Stratified sam- pling meets machine learning,

    E. Liberty, Z. Karnin, B. Xiang, L. Rouesnel, B. Coskun, R. Nallapati, J. Delgado, A. Sadoughi, Y . Astashonok, P. Daset al., “Stratified sam- pling meets machine learning,” inProceedings of The 33rd International Conference on Machine Learning, ser. Proceedings of Machine Lea...

    2016

  30. [38]

    Exploratory data analysis,

    J. W. Tukey, “Exploratory data analysis,” 1977

    1977

  31. [39]

    Protecting privacy when disclosing in- formation: k-anonymity and its enforcement through generalization and suppression,

    P. Samarati and L. Sweeney, “Protecting privacy when disclosing in- formation: k-anonymity and its enforcement through generalization and suppression,”Technical report, SRI International, 1998

    1998

  32. [40]

    Mnist handwritten digit database,

    Y . LeCun and C. Cortes, “Mnist handwritten digit database,” 2010. [Online]. Available: http://yann.lecun.com/exdb/mnist/

    2010

  33. [41]

    Learning multiple layers of features from tiny images,

    A. Krizhevsky, “Learning multiple layers of features from tiny images,” Tech. Rep., 2009

    2009

  34. [42]

    Flamby: Datasets and bench- marks for cross-silo federated learning in realistic healthcare settings,

    C. He, S. Rasouli, I. Zachariah, P. Tiwari, P. Bacon, Y . Shen, A. Kotti, O. Marfoq, H. Benali, T. Clozelet al., “Flamby: Datasets and bench- marks for cross-silo federated learning in realistic healthcare settings,” arXiv preprint arXiv:2210.04620, 2022

    arXiv

  35. [43]

    Flagship dataset of type 2 diabetes from the ai-readi project (1.0.0),

    A.-R. Consortium, “Flagship dataset of type 2 diabetes from the ai-readi project (1.0.0),” 2024. [Online]. Available: https://doi.org/10. 60775/fairhub.1

    2024

  36. [44]

    Delta: Balancing gpu performance with advanced system interfaces,

    W. Gropp, T. Boerner, B. Bode, and G. Bauer, “Delta: Balancing gpu performance with advanced system interfaces,” National Center for Su- percomputing Applications, University of Illinois at Urbana-Champaign, Technical Report, 2023, funded by National Science Foundation (award ...

    2023

  37. [45]

    FLamby IXI Dataset,

    Owkin, “FLamby IXI Dataset,” https://github.com/owkin/FLamby/blob/ main/flamby/datasets/fed ixi/README.md#prediction-task, 2021, ac- cessed: 2025-07-14

    2021

  38. [46]

    An experimental study of class imbalance in federated learning,

    C. Xiao and S. Wang, “An experimental study of class imbalance in federated learning,”arXiv preprint arXiv:2109.04094, 2022

    2022 arXiv

  39. [47]

    R. Labs. (2023) Understanding the impact of class imbalance in federated learning. [Online]. Available: https://risingwave.com/blog/ understanding-the-impact-of-class-imbalance-in-federated-learning/

    2023

  40. [48]

    Fairness and accuracy in federated learning,

    W. Huang, T. Li, D. Wang, S. Du, and J. Zhang, “Fairness and accuracy in federated learning,”Information Sciences, vol. 589, pp. 170–185, 2022

    2022

  41. [49]

    Partially-federated learning: A new approach to achieving privacy and effectiveness,

    M. Fisichella, G. Lax, and A. Russo, “Partially-federated learning: A new approach to achieving privacy and effectiveness,”Information Sciences, vol. 610, pp. 1–18, 2022

    2022

  42. [50]

    Fedmef: Towards memory-efficient federated dynamic pruning,

    H. Huang, W. Zhuang, C. Chen, and L. Lyu, “Fedmef: Towards memory-efficient federated dynamic pruning,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Computer Vision Foundation, 2024

    2024

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.