A New One-Shot Federated Learning Framework for Medical Imaging Classification with Feature-Guided Rectified Flow and Knowledge Distillation

Guibo Luo; Hanwen Zhang; Qiya Yang; Yuesheng Zhu; Yufei Ma

REVIEW 5 major objections 5 minor 43 references

A New One-Shot Federated Learning Framework for Medical Imaging Classification with Feature-Guided Rectified Flow and Knowledge Distillation

T0 review · 5 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read The paper claims that one-shot federated learning can beat multi-round federated learning on medical image classification when clients share a feature-level rectified flow model and the server trains the global model with dual-layer…

desk verdict Solid engineering result with a plausible accuracy story; the privacy claim is the part that needs real work before I'd trust it. read the letter →

arxiv 2507.19045 v2 pith:BRCKTL6T submitted 2025-07-25 cs.CV cs.DC

Yufei Ma , Hanwen Zhang , Qiya Yang , Guibo Luo , Yuesheng Zhu This is my paper · ORCID

classification cs.CVcs.DC

keywords one-shotfederatedlearningrectifiedflowknowledgedistillationmedicalimageclassificationnon-IIDdataprivacy-preservinggenerationfeature-levelsynthesisdata-freeaggregation

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 one-shot federated learning can beat multi-round federated learning for medical image classification if clients upload a feature-level generative model and the server trains the global model with dual-layer knowledge distillation. Clients train a convolutional plus tanh feature extractor paired with a classifier, then a rectified flow model on the extracted features; only these trained models are transmitted. The server samples synthetic features from every client, forms an averaged teacher from client classifiers, and trains a student on synthetic features with cross-entropy, logit distillation, and intermediate-feature alignment. On three non-IID medical datasets (brain-tumor MRI, chest X-ray tuberculosis, skin-lesion dermoscopy) the method reaches 60.15%, 86.32%, and 70.01% accuracy, outperforming the tested multi-round federated learning methods and a medical-imaging one-shot baseline. If correct, this would make single-round aggregation a practical and more private option for multi-center medical imaging.

What carries the argument

Feature-Guided Rectified Flow (FG-RF) plus Dual-Layer Knowledge Distillation (DLKD). In FG-RF, each client trains a lossy feature extractor (a convolutional layer followed by tanh) and a classifier; the rectified flow model learns the velocity field $g_\phi(z_t, t)$ that transports Gaussian noise $z_T$ to target features $z_0$ along a linear interpolation path, and the server samples synthetic features $\hat{z}$ by solving the ODE. In DLKD, the server averages uploaded classifiers into a teacher $\bar{c}_\gamma$ and trains a student $s_\psi$ on synthetic features with $(1-\alpha)$ cross-entropy, $\alpha$ KL divergence against teacher logits, and $\beta$ $\ell_2$ alignment of the student's third-layer features to the teacher's; this two-depth alignment is the mechanism claimed to reduce both the synthetic-real distribution gap and the logits-only distillation error.

What would settle it

Train a powerful, unconstrained reconstruction decoder against the synthetic features produced by FG-RF and measure the minimum $\ell_2$ distance between reconstructed and original training images; if any original image is recovered at or below the paper's own 0.1 memorization threshold, the claimed strict privacy advantage of feature-level generation over pixel-level generation fails.

Watch

Extended reading notes

Core claim

The central claim is that single-round aggregation can converge under non-IID medical data if the synthetic data are generated in a lossy feature space rather than pixel space, and if the student model imitates teachers at two depths. The paper shows this concretely: on three datasets, the framework exceeds the accuracy of multi-round federated learning baselines and of a comparable one-shot medical baseline, while training and inference are far faster than a pixel-level DDPM. Theoretically, the authors argue from the data-processing inequality that a Markov chain $x \to z \to \hat{z}$ obeys $I(x; \hat{z}) \leq I(x; z) < I(x; \hat{x})$, so feature-level generation leaks less about the original image than pixel-level generation, and from an $H$-divergence bound that aligning intermediate features reduces both the distribution gap and the distillation error. Experimentally, feature-level synthetic images have larger $\ell_2$ distance from original images under a memorization threshold, which the paper interprets as evidence of reduced privacy leakage.

Load-bearing premise

The privacy guarantee rests on the claim that the conv+tanh feature extractor is truly lossy — that it destroys enough image information so no decoder can reconstruct the original image; the paper asserts this lossiness but does not measure or bound it for the actual extractor.

Editorial extensions

If this is right

A single communication round is enough to match or exceed 100-round federated learning on the three datasets tested, so communication costs drop by roughly two orders of magnitude.
Feature-level rectified flow is much faster than pixel-level DDPM: training time roughly halves and per-image inference drops from tens of seconds to under a second on the reported datasets.
Feature-level synthetic images stay farther from original images under the $\ell_2$ memorization metric used, giving the method a measurable privacy margin over pixel-level generation.
The intermediate-feature alignment in DLKD is load-bearing: removing it lowers accuracy, for example from 86.32% to 84.74% on chest X-ray.
The method reproduces across MRI, chest X-ray, and dermoscopy, indicating the framework is not restricted to a single imaging modality.

Reading between the lines

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

A natural next step would be to measure mutual information or reconstruction error for the actual conv+tanh extractor on each dataset; a numeric bound would convert the qualitative privacy argument into a testable guarantee.
Since DLKD only requires a chosen intermediate layer in student and teacher, the same scheme could in principle aggregate clients with different backbone architectures once a layer correspondence is fixed; the paper does not test this.
The reported speedups suggest that the efficiency advantage of feature-level rectified flow will grow with image resolution, making the approach a plausible fit for higher-resolution or volumetric medical scans.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note

read the letter

Quick take: the paper is worth a serious look for the OSFL crowd. The new combination is feature-level rectified flow for one-shot federated learning plus dual-layer knowledge distillation. That specific packaging isn't in the prior work they cite, and the efficiency numbers are credible: FG-RF trains and samples much faster than the DDPM baseline. The accuracy results on three medical datasets are also suggestive—beating multi-round FL with one round is the kind of thing people in this subfield care about, and the ablations show each added component helps.

The soft spots are real but not all equal. The abstract numbers don't match Table 1: the abstract claims up to 21.73% improvement over multi-round FL and an average 21.75% over FedISCA, but the text reports 31.44% over FedAvg on Chest X-Ray and 48.51% over FedISCA on HAM10000. That's a glaring internal inconsistency; it should be fixed before anything else. Missing error bars and a single seed per experiment make the accuracy claims weaker than they look. I'd want at least three seeds for the main table.

The privacy argument is the load-bearing weak spot. Section 3.4 assumes the feature extractor is lossy and concludes H(x|z)>0, but the extractor is a stride-1 conv plus tanh—tanh is bijective and the conv can be near-invertible. They never measure H(x|z) or bound reconstruction error against a real adversary. The DPI chain is also shaky because the RFM is label-conditioned, so the Markov assumption isn't guaranteed. And the Table 5 test uses a decoder they trained, i.e., a cooperative actor, not a worst-case adversary. I don't think the privacy conclusion is salvageable as stated. The accuracy results are separable, so the paper could stand without the privacy proof if they reframe that section as a heuristic argument.

The generalization bound is more of a repackaging of Ben-David plus a claim that DLKD reduces both terms—no real novelty, but it doesn't hurt. The self-citations are partly from the same group but mostly to prior OSFL work in the area, not a red flag.

Who this is for: people working on one-shot federated learning or medical FL. They'll get a concrete method with code and a dataset setup they can build on. It deserves peer review, but only after the authors fix the number mismatch, add variance stats, and weaken the privacy claims to match the evidence. I'd engage with it.

Referee Report

5 major / 5 minor

Summary. This paper proposes a one-shot federated learning framework for medical image classification. Each client trains a feature extractor and classifier on its local non-IID data, then trains a rectified flow model on the extracted features, and uploads only the classifier and flow model to the server. The server samples synthetic features from the flow models, aggregates the client classifiers into a teacher ensemble, and trains a global student via Dual-Layer Knowledge Distillation that matches both output logits and intermediate-layer features. Experiments on three medical imaging datasets report accuracies of 60.15%, 86.32%, and 70.01%, which the authors interpret as improvements over multi-round FL, over other one-shot methods, and over a DDPM variant, along with lower training and inference times. The paper also presents an information-theoretic privacy analysis and a generalization bound for the distillation step.

Significance. If the empirical results are reproducible, the framework is a useful contribution: it targets a practical problem, one-shot federated learning under non-IID medical data, with a concrete mechanism combining feature-level rectified flow synthesis and two-level knowledge distillation. The reported accuracy gains over FedISCA and over multi-round FL are large, the efficiency comparison with DDPM is favorable, and the authors release code. However, the privacy guarantee, which is a headline contribution, is not established by the current argument, and the empirical claim is undermined by missing variance/seed reporting and by inconsistencies between the abstract and Table 1. The accuracy result is separable from the privacy claim and could stand after revision, but the paper needs substantial reworking before the central claims are supported.

major comments (5)

[Section 3.4, Eqs. (12)-(16)] The data-processing inequality step I(x;zhat) <= I(x;z) is not valid for the proposed pipeline because the rectified flow model is trained and sampled conditionally on the class label y (Figure 4 and Algorithm 1 line 15), and y is not a deterministic function of z; zhat can therefore carry information about x through y that is not mediated by z, so the Markov chain x -> z -> zhat is not guaranteed. In addition, the assumption H(x|z)>0 in Eq. (15) is asserted rather than established: the feature extractor is a stride-1 convolution followed by tanh, and tanh is bijective while a stride-1 convolution can be near-invertible. Please either measure or bound the mutual information for the actual extractor, or replace the formal privacy claim with an empirical reconstruction study under a well-defined threat model.
[Section 3.4, Eq. (14), and Section 4.5, Table 5] Equation (14) assumes that pixel-level generation achieves H(x|xhat) ≈ 0, but no memorization test is provided to support this, and generative models trained on finite data do not in general memorize every training example. The empirical check in Table 5 also does not establish a privacy advantage: the authors train a feature decoder themselves, which is a cooperative actor rather than a worst-case adversary, and the threshold delta=0.1 from reference [5] is applied without explaining why it remains meaningful for decoded feature-space outputs. Larger average L2 distances under this protocol do not show that feature-level generation leaks less information than pixel-level generation. Please state a concrete adversary model and report reconstruction success or information leakage under that model.
[Abstract and Section 4.2, Table 1] The central empirical claim is not consistently reported. The abstract states an improvement of up to 21.73% and an average improvement over FedISCA of 21.75%, but the numbers in Table 1 imply improvements over FedAvg of 23.20, 31.44, and 14.93 percentage points (up to 31.44) and improvements over FedISCA of 23.38, 32.78, and 48.51 percentage points (average 34.89). Table 1 also reports no standard deviations, no number of seeds, and no statistical significance. Please correct the abstract, report mean and standard deviation over multiple seeds, and describe the evaluation protocol precisely.
[Section 3.4, Eqs. (17)-(21)] The generalization analysis is a standard domain-adaptation bound, and the claim that DLKD reduces both terms is not derived. The feature-alignment loss beta L_feat is minimized on synthetic features, but no argument shows that it reduces the H-divergence d_H(D_real, D_syn) or the distillation consistency term |eps_gen(s_psi) - eps_real(cbar_gamma)|. The symbol lambda is also reused for the loss weight in Eq. (1) and for the minimum combined error in Eq. (20), which is confusing. Please either provide a proof of the claimed reduction or state the bound as motivation and remove the claim of proof.
[Section 4.1 and Algorithm 1] The experimental setup is underspecified to the point of preventing replication. The paper does not state the train/test split, the evaluation protocol, the client/server hyperparameters (learning rates, numbers of epochs, alpha, beta, temperature T, and selected intermediate layer l), or the exact non-IID sharding procedure for the Brain Tumor and HAM10000 datasets. It is also unclear how a client uses the returned student model s_psi on raw images, since s_psi is trained on feature vectors while the feature extractor is not uploaded to the server. The lack of this information, together with the absence of error bars, makes it difficult to rule out data leakage or test-set contamination.

minor comments (5)

[Section 3 heading and Section 3.3] The acronym is spelled inconsistently: Section 3 introduces 'Dual-Layer Knowledge Distillation (DLDK)', while the title, abstract, and Algorithm 1 use DLKD; please standardize.
[Eq. (1) and Algorithm 1] Equation (1) writes the cross-entropy loss as L_CE(s_psi(x), y) even though D_syn is composed of synthetic features zhat; the student model input should be written consistently as z or zhat.
[Algorithm 1, line 7] The notation z_i,T and z_i,0 is introduced without defining the noise endpoint, and T is also used for the total number of epochs; please rename one of these quantities to avoid ambiguity.
[Table 4] The ablation row labeled 'w/o DLKD and feature extractor' is not clearly defined; please state which components are removed in each row, especially whether this row corresponds to pixel-level generation.
[Section 4.5, Table 5] The phrases 'Train with pictures' and 'Train with features' are unclear; the caption should describe the two conditions, the decoder training procedure, and why delta=0.1 applies to decoded features.

Circularity Check

1 steps flagged · score 4.0 of 10

Accuracy claims are empirically self-contained, but the §3.4 privacy guarantee is self-definitional: it assumes the feature extractor is lossy (H(x|z)>0) and then 'proves' feature-level generation leaks less, so the theoretical privacy advantage reduces to its construction assumption.

self definitional [Section 3.4, Eqs. (12)-(16); supporting design claim in Section 3.2 ('Feature Extractor Design')]
"Given that the original data x is transformed into a feature z through an irreversible and lossy feature extractor fθ ... the transformation x→z inevitably leads to information loss due to the lossy nature of the feature extractor fθ, i.e., H(x|z) > 0. ... This indicates that the mutual information between x and the generated feature ˆz is strictly less than that between x and ˆx in pixel-level generation, suggesting a lower risk of privacy leakage in feature-level generation."

The inequality I(x;ˆz) < I(x;ˆx) is derived from H(x|z)>0, which is asserted as an inevitable property of the feature extractor rather than measured. That same lossiness is the paper's stated design rationale for the feature extractor, which is chosen because it 'irreversibly perturbs the original image, thereby preventing excessive information leakage.' Thus the theorem's conclusion is entailed by its own construction assumption: if the extractor were not lossy, Eq. (15) would fail. The privacy guarantee is a restatement of the design premise, not an independent derivation. The empirical Table 5 test is a separate, weaker check using a cooperative decoder, but the theoretical proof itself is circular.

full rationale

The main accuracy derivation is not circular. The framework trains local feature extractors and rectified flow models, synthesizes a server-side feature dataset, and applies dual-layer knowledge distillation; the reported gains over FedAvg, FedISCA, and a DDPM baseline are empirical comparisons with ablations, and no fitted parameter is renamed as a prediction. The self-citations (e.g., refs. 24-26, 35, 39, 43) are used as background, dataset sources, or prior OSFL/medical diffusion work and are not load-bearing for the central empirical claim. The one genuine circular step is the privacy analysis in Section 3.4: Eq. (15) concludes I(x;z)<H(x) solely from the assertion that the feature extractor is 'lossy' and 'irreversible,' which is exactly the design property used to motivate the extractor. The theorem therefore proves the privacy advantage by assuming it. Separately, the DPI chain x→z→ˆz is not established because the rectified flow model is label-conditioned during training (Figure 4 and Algorithm 1), so ˆz may carry information about x beyond z; this is a correctness gap rather than a circularity. Since the accuracy contribution remains independent and empirically supported, the overall circularity is partial and localized to the theoretical privacy claim.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The central empirical claim rests on standard learning theory and the unverified assumption that the feature extractor discards information. The DLKD objective introduces two weighting hyperparameters and a layer choice that are not reported. No new physical or conceptual entities are postulated.

free parameters (5)

loss weight alpha = not reported
Balances cross-entropy and KL divergence in Eq. (11); value must be chosen on a validation set.
loss weight beta = not reported
Weights the intermediate feature alignment loss in Eq. (11).
distillation temperature T = not reported
Controls softmax smoothness in Eq. (9).
selected intermediate layer l = third layer of ResNet-18
Choice of layer for feature alignment; could affect results.
privacy threshold delta = 0.1
Borrowed from Carlini et al. to define memorization; its meaning for reconstructed features is unclear.

assumptions (4)

standard math Shannon mutual information and the data processing inequality hold for the defined Markov chain x -> z -> z_hat.
Used in Eq. (12) and (16) to bound information leakage.
domain assumption The feature extractor is lossy, so H(x|z) > 0.
Central to the privacy proof in Section 3.4; never measured.
standard math Ben-David domain adaptation bound applies to the synthetic and real distributions.
Used in Eq. (20) to bound the generalization gap.
domain assumption For pixel-level generation, the conditional entropy H(x|x_hat) is approximately zero.
Needed for Eq. (14); depends on the model fully memorizing the data.

how reviews work

0 comments

Cite this review

Pith. "Pith review of A New One-Shot Federated Learning Framework for Medical Imaging Classification with Feature-Guided Rectified Flow and Knowledge Distillation." pith.science (2026). https://pith.science/paper/BRCKTL6T

@misc{pith2026250719045,
  author       = {Pith},
  title        = {Pith review of: A New One-Shot Federated Learning Framework for Medical Imaging Classification with Feature-Guided Rectified Flow and Knowledge Distillation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/BRCKTL6T}},
  note         = {Machine review of arXiv:2507.19045}
}

read the original abstract

In multi-center scenarios, One-Shot Federated Learning (OSFL) has attracted increasing attention due to its low communication overhead, requiring only a single round of transmission. However, existing generative model-based OSFL methods suffer from low training efficiency and potential privacy leakage in the healthcare domain. Additionally, achieving convergence within a single round of model aggregation is challenging under non-Independent and Identically Distributed (non-IID) data. To address these challenges, in this paper a modified OSFL framework is proposed, in which a new Feature-Guided Rectified Flow Model (FG-RF) and Dual-Layer Knowledge Distillation (DLKD) aggregation method are developed. FG-RF on the client side accelerates generative modeling in medical imaging scenarios while preserving privacy by synthesizing feature-level images rather than pixel-level images. To handle non-IID distributions, DLKD enables the global student model to simultaneously mimic the output logits and align the intermediate-layer features of client-side teacher models during aggregation. Experimental results on three non-IID medical imaging datasets show that our new framework and method outperform multi-round federated learning approaches, achieving up to 21.73% improvement, and exceeds the baseline FedISCA by an average of 21.75%. Furthermore, our experiments demonstrate that feature-level synthetic images significantly reduce privacy leakage risks compared to pixel-level synthetic images. The code is available at https://github.com/LMIAPC/one-shot-fl-medical.

Figures

Figures reproduced from arXiv: 2507.19045 by the authors.

**Figure 1.** t-SNE visualization of the TB dataset before and after training without knowledge distillation. Left: initial feature distribution shows significant client discrepancy (non-IID). Right: after one communication round, features remain scattered, indicating insufficient convergence due to local distribution divergence. To address the challenges of generative model-based OSFL for medical image classification, we propose… view at source ↗

**Figure 2.** The overall framework of our proposed method. 3.1 Problem formulation Assuming that we have K clients with their private dataset DK = {(xi, yi)} Nk i=1, k = 1, . . . , K, where xi represents the image, yi represents the class label, and N k is the number of images for the k-th client. In the FG-RF component, each client trains a feature extractor fθk to obtain image features, which are then used to train a local R… view at source ↗

**Figure 3.** Feature extractor training diagram. FG-RF Training. After obtaining the features zk extracted from the original images using the feature extractor, each client trains its respective RFM. Unlike DDPM, the RFM models a smooth and deterministic trajectory that directly maps the randomly initialized features zk,T to the target features zk,0 via an ordinary differential equation (ODE): dzk dt = gϕk (zk,t, t) (3) here,… view at source ↗

Figures from the paper (2 more)

**Figure 5.** Figure 5: t-SNE visualization of the TB dataset before and after training with DLKD. merged after DLKD training, implying that DLKD significantly reduces feature distribution discrepancies across clients. Each client first trains a teacher classifier cγk on its real data. The i…

**Figure 4.** Figure 4: The Rectified Flow Model training diagram. Features Sampling. After completing the FG-RF training, each client will send its trained RFM model gϕk to the server. The server then uses the gϕk trained by each client to infer the image features zˆk. The sampling process i…

Review history (2 revisions) →

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

43 extracted references · 25 canonical work pages

[5]

Carlini, J

N. Carlini, J. Hayes, M. Nasr, M. Jagielski, V . Sehwag, F. Tramer, B. Balle, D. Ippolito, and E. Wallace. Extracting training data from diffusion models. In 32nd USENIX Security Symposium (USENIX Se- curity 23), pages 5253–5270, 2023

work page 2023
[1]

https://www.kaggle

Brain Tumor Classification (MRI) — kaggle.com. https://www.kaggle. com/datasets/sartajbhuvaji/brain-tumor-classification-mri. [Accessed 30-04-2025]

work page 2025
[2]

Beitollahi, A

M. Beitollahi, A. Bie, S. Hemati, L. M. Brunswic, X. Li, X. Chen, and G. Zhang. Parametric feature transfer: One-shot federated learning with foundation models. arXiv preprint arXiv:2402.01862, 2024

arXiv 2024
[3]

Ben-David, J

S. Ben-David, J. Blitzer, K. Crammer, A. Kulesza, F. Pereira, and J. W. Vaughan. A theory of learning from different domains. Machine learn- ing, 79:151–175, 2010

work page 2010
[4]

Bloch, O

M. Bloch, O. Günlü, A. Yener, F. Oggier, H. V . Poor, L. Sankar, and R. F. Schaefer. An overview of information-theoretic security and privacy: Metrics, limits and applications. IEEE Journal on Selected Areas in Information Theory, 2(1):5–22, 2021

work page 2021
[6]

H. Chen, H. Li, Y . Zhang, J. Bi, G. Zhang, Y . Zhang, P. Torr, J. Gu, D. Krompass, and V . Tresp. Fedbip: Heterogeneous one-shot feder- ated learning with personalized latent diffusion models. arXiv preprint arXiv:2410.04810, 2024

arXiv
[7]

R. Dai, Y . Zhang, A. Li, T. Liu, X. Yang, and B. Han. Enhancing one- shot federated learning through data and ensemble co-boosting. arXiv preprint arXiv:2402.15070, 2024

arXiv 2024
[8]

Y . Deng, Y . Zhou, G. Liu, J. H. Wang, and Y . Shui. Enhancing fed- erated learning by one-shot transferring of intermediate features from clients. In 2023 IEEE 10th International Conference on Data Science and Advanced Analytics (DSAA), pages 1–11. IEEE, 2023

work page 2023

Show all 43 references

[9]

N. Guha, A. Talwalkar, and V . Smith. One-shot federated learning. arXiv preprint arXiv:1902.11175, 2019

1902 arXiv
[10]

K. He, X. Zhang, S. Ren, and J. Sun. Deep residual learning for image recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 770–778, 2016

2016
[11]

C. E. Heinbaugh, E. Luz-Ricca, and H. Shao. Data-free one-shot feder- ated learning under very high statistical heterogeneity. In The Eleventh International Conference on Learning Representations, 2023

2023
[12]

Hinton, O

G. Hinton, O. Vinyals, and J. Dean. Distilling the knowledge in a neural network. arXiv preprint arXiv:1503.02531, 2015

2015 arXiv
[13]

J. Ho, A. Jain, and P. Abbeel. Denoising diffusion probabilistic mod- els. Advances in neural information processing systems, 33:6840–6851, 2020

2020
[14]

M. Kang, P. Chikontwe, S. Kim, K. H. Jin, E. Adeli, K. M. Pohl, and S. H. Park. One-shot federated learning on medical data using knowl- edge distillation with image synthesis and client model adaptation. In International Conference on Medical Image Computing and Computer- Ass...

2023
[15]

S. P. Karimireddy, S. Kale, M. Mohri, S. Reddi, S. Stich, and A. T. Suresh. Scaffold: Stochastic controlled averaging for federated learn- ing. In International conference on machine learning , pages 5132–
[16]

Kasturi and C

A. Kasturi and C. Hota. Osgan: One-shot distributed learning using generative adversarial networks. The Journal of Supercomputing , 79 (12):13620–13640, 2023

2023
[17]

Kasturi, A

A. Kasturi, A. R. Ellore, and C. Hota. Fusion learning: A one shot fed- erated learning. In Computational Science–ICCS 2020: 20th Interna- tional Conference, Amsterdam, The Netherlands, June 3–5, 2020, Pro- ceedings, Part III 20, pages 424–436. Springer, 2020

2020
[18]

Kolchinsky, B

A. Kolchinsky, B. D. Tracey, and D. H. Wolpert. Nonlinear information bottleneck. Entropy, 21(12):1181, 2019

2019
[19]

Q. Li, B. He, and D. Song. Practical one-shot federated learning for cross-silo setting. arXiv preprint arXiv:2010.01017, 2020

2010 arXiv
[20]

Q. Li, B. He, and D. Song. Model-contrastive federated learning. In Proceedings of the IEEE/CVF conference on computer vision and pat- tern recognition, pages 10713–10722, 2021

2021
[21]

T. Li, A. K. Sahu, M. Zaheer, M. Sanjabi, A. Talwalkar, and V . Smith. Federated optimization in heterogeneous networks. Proceedings of Ma- chine learning and systems, 2:429–450, 2020

2020
[22]

S. Liu, R. Zhang, M. Fang, H. Li, T. Xun, Z. Wang, W. Shang, J. Tian, and D. Dong. Pcrfed: personalized federated learning with contrastive representation for non-independently and identically distributed medi- cal image segmentation. Visual Computing for Industry, Biomedicine...

2025
[23]

X. Liu, C. Gong, and Q. Liu. Flow straight and fast: Learning to generate and transfer data with rectified flow. arXiv preprint arXiv:2209.03003, 2022

2022 arXiv
[24]

Y . Liu, G. Luo, and Y . Zhu. Fedfms: Exploring federated founda- tion models for medical image segmentation. In International Con- ference on Medical Image Computing and Computer-Assisted Interven- tion, pages 283–293. Springer, 2024

2024
[25]

G. Luo, T. Liu, J. Lu, X. Chen, L. Yu, J. Wu, D. Z. Chen, and W. Cai. In- fluence of data distribution on federated learning performance in tumor segmentation. Radiology: Artificial Intelligence, 5(3):e220082, 2023

2023
[26]

G. Luo, H. Zhang, X. Wang, M. Chen, and Y . Zhu. Mpcpa: Multi-center privacy computing with predictions aggregation based on denoising dif- fusion probabilistic model. arXiv preprint arXiv:2403.07838, 2024

2024 arXiv
[27]

Luo and S

J. Luo and S. Wu. Fedsld: Federated learning with shared label distribu- tion for medical image classification. In 2022 IEEE 19th International Symposium on Biomedical Imaging (ISBI), pages 1–5. IEEE, 2022

2022
[28]

A new one-shot federated learning framework for medical imaging classification with feature-guided rectified flow and knowl- edge distillation

Y . Ma, H. Zhang, Q. Yang, G. Luo, and Y . Zhu. Supplementary ma- terial for "A new one-shot federated learning framework for medical imaging classification with feature-guided rectified flow and knowl- edge distillation". Zenodo, 2025. Available at https://doi.org/10.5281/ ze...

2025
[29]

McMahan, E

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

2017
[30]

Peebles and S

W. Peebles and S. Xie. Scalable diffusion models with transformers. In Proceedings of the IEEE/CVF international conference on computer vision, pages 4195–4205, 2023

2023
[31]

Rombach, A

R. Rombach, A. Blattmann, D. Lorenz, P. Esser, and B. Ommer. High- resolution image synthesis with latent diffusion models. InProceedings of the IEEE/CVF conference on computer vision and pattern recogni- tion, pages 10684–10695, 2022

2022
[32]

C. E. Shannon. A mathematical theory of communication. The Bell system technical journal, 27(3):379–423, 1948

1948
[33]

Tschandl, C

P. Tschandl, C. Rosendahl, and H. Kittler. The ham10000 dataset, a large collection of multi-source dermatoscopic images of common pig- mented skin lesions. Scientific data, 5(1):1–9, 2018

2018
[34]

J. Wang, Q. Liu, H. Liang, G. Joshi, and H. V . Poor. Tackling the objective inconsistency problem in heterogeneous federated optimiza- tion. Advances in neural information processing systems , 33:7611– 7623, 2020

2020
[35]

J. Xie, Z. Zhang, Z. Weng, Y . Zhu, and G. Luo. Meddiff-ft: Data- efficient diffusion model fine-tuning with structural guidance for con- trollable medical image synthesis. arXiv preprint arXiv:2507.00377 , 2025

2025 arXiv
[36]

M. Yang, S. Su, B. Li, and X. Xue. One-shot federated learning with classifier-guided diffusion models. arXiv preprint arXiv:2311.08870 , 2023

2023 arXiv
[37]

M. Yang, S. Su, B. Li, and X. Xue. Exploring one-shot semi-supervised federated learning with pre-trained diffusion models. In Proceedings of the AAAI Conference on Artificial Intelligence , volume 38, pages 16325–16333, 2024

2024
[38]

M. Yang, S. Su, B. Li, and X. Xue. Feddeo: Description-enhanced one- shot federated learning with diffusion models. In Proceedings of the 32nd ACM International Conference on Multimedia, pages 6666–6675, 2024

2024
[39]

Zhang, M

H. Zhang, M. Chen, Y . Liu, G. Luo, and Y . Zhu. Non-iid medical im- age segmentation based on cascaded diffusion model for diverse multi- center scenarios. IEEE Journal of Biomedical and Health Informatics , 2025

2025
[40]

Zhang, C

J. Zhang, C. Chen, B. Li, L. Lyu, S. Wu, S. Ding, C. Shen, and C. Wu. Dense: Data-free one-shot federated learning. Advances in Neural In- formation Processing Systems, 35:21414–21428, 2022

2022
[41]

Zhang, X

J. Zhang, X. Qi, and B. Zhao. Federated generative learning with foun- dation models. arXiv preprint arXiv:2306.16064, 2023

2023 arXiv
[42]

Y . Zhou, G. Pu, X. Ma, X. Li, and D. Wu. Distilled one-shot federated learning. arXiv preprint arXiv:2009.07999, 2020

2009 arXiv
[43]

Z. Zhou, G. Luo, M. Chen, Z. Weng, and Y . Zhu. Federated learning for medical image classification: A comprehensive benchmark. arXiv preprint arXiv:2504.05238, 2025

2025 arXiv

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.

[1] [5]

Carlini, J

N. Carlini, J. Hayes, M. Nasr, M. Jagielski, V . Sehwag, F. Tramer, B. Balle, D. Ippolito, and E. Wallace. Extracting training data from diffusion models. In 32nd USENIX Security Symposium (USENIX Se- curity 23), pages 5253–5270, 2023

work page 2023

[2] [1]

https://www.kaggle

Brain Tumor Classification (MRI) — kaggle.com. https://www.kaggle. com/datasets/sartajbhuvaji/brain-tumor-classification-mri. [Accessed 30-04-2025]

work page 2025

[3] [2]

Beitollahi, A

M. Beitollahi, A. Bie, S. Hemati, L. M. Brunswic, X. Li, X. Chen, and G. Zhang. Parametric feature transfer: One-shot federated learning with foundation models. arXiv preprint arXiv:2402.01862, 2024

arXiv 2024

[4] [3]

Ben-David, J

S. Ben-David, J. Blitzer, K. Crammer, A. Kulesza, F. Pereira, and J. W. Vaughan. A theory of learning from different domains. Machine learn- ing, 79:151–175, 2010

work page 2010

[5] [4]

Bloch, O

M. Bloch, O. Günlü, A. Yener, F. Oggier, H. V . Poor, L. Sankar, and R. F. Schaefer. An overview of information-theoretic security and privacy: Metrics, limits and applications. IEEE Journal on Selected Areas in Information Theory, 2(1):5–22, 2021

work page 2021

[6] [6]

H. Chen, H. Li, Y . Zhang, J. Bi, G. Zhang, Y . Zhang, P. Torr, J. Gu, D. Krompass, and V . Tresp. Fedbip: Heterogeneous one-shot feder- ated learning with personalized latent diffusion models. arXiv preprint arXiv:2410.04810, 2024

arXiv

[7] [7]

R. Dai, Y . Zhang, A. Li, T. Liu, X. Yang, and B. Han. Enhancing one- shot federated learning through data and ensemble co-boosting. arXiv preprint arXiv:2402.15070, 2024

arXiv 2024

[8] [8]

Y . Deng, Y . Zhou, G. Liu, J. H. Wang, and Y . Shui. Enhancing fed- erated learning by one-shot transferring of intermediate features from clients. In 2023 IEEE 10th International Conference on Data Science and Advanced Analytics (DSAA), pages 1–11. IEEE, 2023

work page 2023