FedSC: Federated Learning with Semantic-Aware Collaboration

Haoran Li; Huaming Chen; Huan Wang; Jiahua Shi; Jun Shen; Jun Yan

REVIEW 5 major objections 5 minor 77 references

FedSC: Federated Learning with Semantic-Aware Collaboration

T0 review · 5 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read FedSC shares class prototypes across clients and beats nine baselines under heterogeneous data.

desk verdict Sensible FL method with a broken convergence proof; the empirical work is substantial and the method is worth trying, but the theory should be fixed or dropped. read the letter →

arxiv 2506.21012 v1 pith:VIOXQVG4 submitted 2025-06-26 cs.CV

Huan Wang , Haoran Li , Huaming Chen , Jun Yan , Jiahua Shi , Jun Shen This is my paper · ORCID

classification cs.CV

keywords federatedlearningdataheterogeneityprototypecontrastivelabelshiftlong-taileddistributionfew-shotclassificationconvergenceanalysis

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

Federated learning across clients with different label distributions usually produces a global model that drifts because each local model overfits its own skewed data. This paper claims that the fix is to let clients exchange class-level semantic knowledge, not only model weights: each client computes class prototypes, and the server fuses them into relational prototypes by angular similarity and consistent prototypes by discrepancy-weighted aggregation. These prototypes feed two complementary losses, a contrastive loss that pulls samples toward their class's relational prototypes and pushes them away from other classes, and a regularization term that pulls features toward stable consistent prototypes. If correct, FedSC gives a more generalizable global model under label shift, long-tailed imbalance, and few-shot tasks, while converging in fewer rounds than existing federated learning methods.

What carries the argument

The argument runs on two constructed objects. Relational prototypes $r^j_k$ are client-prototype means merged over the top-$M$ clients whose class-$j$ prototypes are closest in angle to the global prototype $g^j$, captured by an adjacency matrix $A^j$; they supply diverse, class-relevant semantic knowledge. Consistent prototypes $o^j$ are weighted sums of relational prototypes with weights $e_k = \mathrm{Sigmoid}(a n_k - b d_k)$, where $d_k$ is the distance between the client's local class distribution and a uniform global distribution; they supply a flat, stable convergence target. The two prototype sets carry the learning through loss terms $L_{RPCL}$, a normalized InfoNCE-style contrastive loss with similarity $s(z_i, r^j_k)$ divided by the average feature distance $U$, and $L_{CPDR}$, a squared distance between the sample feature and $o^j$. The convergence theorems then bound the deviation of the combined objective using smoothness, bounded variance, bounded dissimilarity, and Lipschitz continuity assumptions.

What would settle it

One concrete test: run FedSC under a deliberately constructed extreme label shift, for example each client holds exactly two disjoint classes on CIFAR10, and measure per-class accuracy of the final global model. If tail or absent-class accuracy falls below FedAvg, or if the consistent prototypes $o^j$ drift toward head-class directions, the discrepancy-weighting premise fails. On the theory side, empirically estimate $\mathbb{E}_\zeta[\|\nabla L(w,\zeta)-\nabla L(w)\|^2]$ for the RPCL term with the normalization factor $U$; if the variance grows without bound as $U$ varies or fails Assumption 2, the convergence rate in Theorem 3 does not apply to the actual objective.

Watch

Extended reading notes

Core claim

On its own terms, the paper establishes that exploiting intra-client semantic information via prototypes is enough to mitigate the negative effects of data heterogeneity in federated learning. The central mechanism is a local objective $L = L_{CE} + L_{RPCL} + L_{CPDR}$: relational-prototype contrastive learning ($L_{RPCL}$) treats the relational prototypes $R^j$ of class $j$ as virtual teachers, attracting each query embedding and repelling prototypes of other classes; consistent-prototype discrepancy regularization ($L_{CPDR}$) constrains each sample feature to stay close to the corresponding consistent prototype $o^j$. The paper reports that FedSC "universally outperforms" baselines across CIFAR10, CIFAR100, TinyImageNet, and FC100 under three heterogeneity settings, requires fewer rounds to reach target accuracy, and is robust in long-tailed and few-shot scenarios. It also proves, under four stated assumptions, a monotone-decreasing objective and a guaranteed convergence rate with an explicit learning-rate window.

Load-bearing premise

The guarantee depends on assuming that the combined prototype-based loss is smooth, has unbiased stochastic gradients with bounded variance, has bounded gradient dissimilarity, and that feature extraction is Lipschitz continuous; the paper assumes these properties rather than verifying them for the prototype and contrastive terms, and it further assumes a uniform global class distribution when computing the discrepancy weights.

Editorial extensions

If this is right

Under label shift (Dirichlet and NID2 partitions), FedSC improves final accuracy over nine federated learning baselines on all four tested datasets, with the largest reported gains in the hardest heterogeneity regime.
In long-tailed settings with imbalance ratio $\rho$ up to 200, FedSC holds its advantage, which the paper attributes to discrepancy-weighted consistent prototypes counteracting head-class bias.
In few-shot FC100 benchmarks, FedSC outperforms prior prototype- and contrastive-based methods, indicating that the semantic-level knowledge transfers to low-data regimes.
FedSC reaches target accuracies in substantially fewer rounds than FedAvg, FedProx, FedProto, MOON, FedNTD, FedNH, FedRCL, and FedCDA.
Ablation shows RPCL contributes more than CPDR in most settings, and combining both gives the best average accuracy, supporting the paper's complementarity claim.

Reading between the lines

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

The prototype-communication design suggests a privacy and usefulness trade-off: sharing only class means and aggregated angular relations may leak less than sharing model weights, but the paper does not quantify this, so an adversarial membership-inference test on the exchanged prototypes would be a natural next step.
Because the consistent-prototype weights favor clients with large sample size and low distributional discrepancy, the aggregation rule could double as a client-selection or fairness signal; the paper explores neither, but its own Eq. 11 makes the connection immediate.
The angular neighbor selection treats class semantics as living on a hypersphere, so a direct extension would be to test whether replacing the top-$M$ hard selection with a soft, temperature-weighted angular kernel preserves the reported gains at larger client counts.
The empirical gains under long-tail and few-shot settings suggest the method may also act as a regularizer in a single-client long-tail problem, but that transfer is not tested in the paper.

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

The method is sensible and the empirical results are directionally supportive, but the convergence proof in the appendix doesn't hold up, so the "guaranteed convergence" claim should be treated with skepticism.

What's new: FedSC combines angular-neighbor prototype selection (Eq. 8-9), discrepancy-weighted consistent prototypes (Eq. 10-12), and an InfoNCE-style prototype contrastive loss plus a prototype regularization loss. The specific combination isn't in FedProto, FedNH, FedRCL, or FedCDA. That's a legitimate incremental contribution. The method is specified clearly enough to reimplement, and the authors release code, which is good.

What it does well: The empirical evaluation is broad: label shift, long-tailed, few-shot, across four datasets, with comparisons against recent baselines. The accuracy gains are consistent, around 1-3 points, and the ablations show both RPCL and CPDR contribute. The convergence rates table also shows faster reaching of accuracy thresholds.

Soft spots: The main problem is the convergence proof. The stress-test note is correct: in Eq. (10) of the appendix, the proof treats the scalar similarity s(z_i, r^j) as if it were a vector norm, then bounds a log-sum-exp of similarities by prototype norms without a Lipschitz constant for exp. The normalizer U in Eq. (13) is data-dependent and not covered by the assumptions. More seriously, the step after Eq. (12) replaces the sum of squared gradient norms with E B^2, but Assumption 3 gives an upper bound on each squared norm, not a lower bound. With the negative coefficient in front, you cannot tighten the upper bound that way. Without a lower bound on gradients (or a PL condition), the monotone decrease in Theorem 2 and the rate in Theorem 3 do not follow. So the theory as written doesn't establish what it claims.

Two more minor concerns: no standard deviations or seed counts, which matters when gains are a couple of points; and key hyperparameters (tau, M) appear to be tuned on the same benchmark used for the main comparison. The sensitivity analysis in Fig. 5 helps, but it doesn't fully rule out overfitting to that setting.

Who this is for: groups working on prototype-based FL will find the method useful as a baseline or an idea to build on. The flawed theory should not prevent using the method, but it does mean the paper's theoretical contribution is not credible in its current form.

Recommendation: This deserves a serious referee. The empirical work is substantial, but the paper needs major revisions: fix or remove the convergence theorems, add seed variability, and clarify hyperparameter selection. If the theory can't be repaired, it should be replaced with a more modest statement about empirical behavior.

Referee Report

5 major / 5 minor

Summary. The paper proposes FedSC, a federated learning method that augments standard cross-entropy training with two prototype-based losses: Relational Prototypes Contrastive Learning (RPCL), which pulls instance embeddings toward client-combined class prototypes selected by angular similarity, and Consistent Prototypes Discrepancy Regularization (CPDR), which penalizes distance to globally aggregated prototypes weighted by client distribution discrepancy. The authors claim that this semantic-aware collaboration improves accuracy under label shift, long-tailed imbalance, and few-shot settings, and they provide Theorems 1–3 claiming a convergence guarantee and rate under non-convex objectives. Experiments on CIFAR10, CIFAR100, TinyImageNet, and FC100 compare against several FL baselines and report consistent accuracy gains, with ablations for the two modules and hyperparameters.

Significance. If the empirical claims hold, FedSC is a simple and practically relevant contribution: it combines prototype-based semantic information with contrastive and regularization objectives, and the authors release code and evaluate across several heterogeneity regimes. The conceptual distinction between relational prototypes and consistent prototypes is reasonable, and the ablation study suggests both modules contribute. However, the paper's stated theoretical contribution—a guaranteed convergence rate—is not established by the provided proofs, and the experimental reporting lacks statistical uncertainty measures. The empirical headline is directionally supported, but the convergence guarantee is a load-bearing part of the paper's claims and is currently unsupported.

major comments (5)

[Appendix A.3.1, Eq. (10)] The step labeled (ii) is mathematically invalid: it treats the scalar similarity s(z_i, r_j) from Eq. (13) as if it were comparable to the vector r_j, writing s(x,r_j) <= r_j and then bounding log-sum-exp differences by differences of prototype norms. A scalar similarity cannot be bounded by a vector norm in this way, and the log-sum-exp function is not controlled by the proposed inequality without Lipschitz constants for the exponential. The normalizer U in Eq. (13) is data-dependent and is not covered by Assumptions 1–4, so the round-to-round change of L_RPCL is not established.
[Appendix A.3.1, final step of Eq. (13); Appendix A.3.2, Eq. (14)] Assumption 3 states E[||∇L(w,zeta)||^2] <= B, which is an upper bound on the squared gradient norm. In Eq. (13) the coefficient (L1/2 eta^2 - eta) is negative for the small learning rates considered, so replacing the sum of squared gradient norms by E B^2 reverses the inequality: one would need a lower bound on gradient norms, or a Polyak–Lojasiewicz condition, to obtain the claimed negative drift. Consequently, the monotone decrease in Theorem 2 and the rate in Theorem 3 do not follow from the provided analysis.
[Section 4.3 and Appendix A.2] The convergence analysis assumes L = L_CE + L_RPCL + L_CPDR satisfies smoothness, bounded variance, bounded stochastic gradients, and Lipschitz continuity of the feature extractor, but the paper never verifies these properties for the prototype-based losses. The similarity in Eq. (13) involves the data-dependent normalizer U and a softmax over relational prototypes, and the CPDR term depends on consistent prototypes formed by discrepancy-weighted aggregation; none of these are shown to satisfy the assumptions. Theorems 1–3 therefore rely on unverified assumptions.
[Tables 1–3 and Fig. 5] The central accuracy claim is supported only by point estimates. No standard deviations or seed counts are reported, and the hyperparameters M and tau are tuned on the same CIFAR10 NID1 0.2 configuration used for the headline results, which makes the statement that FedSC 'universally outperforms' other methods stronger than the evidence supports.
[Section 3.4, Eq. (10)] The consistent prototypes assume a hypothetical global category distribution H that is uniform over classes. In long-tailed scenarios this is an idealization that may bias the discrepancy weights and the resulting prototypes; the paper does not discuss how sensitive the results are to this choice or whether a non-uniform H would change the conclusions.

minor comments (5)

[Eq. (13)] The displayed formula for the similarity s(z_i, r_j) contains malformed spacing and an unclear decomposition; please rewrite it so that the normalization factor U and the cosine similarity are explicit.
[Eq. (15)] The CPDR loss is written as a sum of squared differences; the notation is fine, but the paper should state explicitly whether this is a squared L2 distance (which is what Eq. (15) computes) rather than an unsquared distance.
[Table 5] The comparison of neighbor-selection metrics reports only final accuracies without variance; adding repeated-run statistics would strengthen the claim that angle-wise selection is superior.
[Algorithm 1, line 8] The global model aggregation uses n_k/N weights over the selected client subset, but N is defined as the total number of samples among all clients; if the subset is not representative, the aggregation weights should be clarified.
[Section 5.2] The sentence 'FedSC universally outperforms other methods' is a summary of the tables, but the evidence is limited to the particular hyperparameter configuration and datasets studied; please temper the wording or provide statistical support.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: FedSC's prototype construction and losses are self-contained, and the convergence proof's apparent defects are correctness concerns rather than circular reasoning.

full rationale

The paper's derivation chain is not circular. Relational prototypes (Eq. 9) are defined as averages of selected clients' class-wise feature means, and consistent prototypes (Eq. 12) are defined as discrepancy-weighted averages of relational prototypes; the RPCL and CPDR losses (Eqs. 14 and 15) are then defined directly from these objects. No parameter is fitted to the reported accuracy and then re-reported as a prediction, and no load-bearing result is imported from the authors' prior work: the only author self-citation ([46], used to mention FedLC in related work) is not load-bearing. The convergence theorems rest on stated Assumptions 1-4 rather than assuming the conclusion; the fact that CPDR uses consistent prototypes as a regularizer is a design choice, not a definitional equivalence. The proof as written does contain apparent formal defects (Assumption 3 is used as a lower bound on squared gradient norms, and Eq. (10) mixes scalar similarities with prototype vectors), but those are correctness concerns about the proof, not circularity in the derivation chain. The empirical claims are benchmarked against external methods and baselines, so the central result has independent content.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

The central method depends on two tuned hyperparameters, tau and M, and on standard but unverified smoothness and boundedness assumptions plus a uniform global distribution idealization. No new physical entities are introduced; relational and consistent prototypes are averages of client features.

free parameters (2)

tau (temperature in RPCL, Eq.14) = 0.05
Controls the strength of contrastive separation; selected by sweeping on CIFAR10 NID1 0.2 test accuracy in Fig.5.
M (number of neighbor clients for relational prototypes, Eq.9) = 2
Number of angular neighbors merged per class; selected on CIFAR10 NID1 0.2 in Fig.5.

assumptions (4)

domain assumption The combined loss L = L_CE + L_RPCL + L_CPDR is L1-smooth, has unbiased stochastic gradients with bounded variance, bounded dissimilarity, and the feature extractor is L2-Lipschitz (Assumptions 1-4, Appendix A.2).
Standard FL assumptions invoked for Theorems 1-3; not verified for the prototype contrastive and regularization losses.
ad hoc to paper The hypothetical global category distribution H is uniform over classes (Eq.10).
Used to define client discrepancy weights for consistent prototypes; is an idealization in long-tailed or strongly biased data.
domain assumption Angular differences between prototypes capture semantic similarity better than Euclidean or cosine distances (Sec.3.3).
The relational prototype neighbor selection depends on this; supported only by one ablation table (Table 5).
domain assumption Averaged prototypes do not leak sensitive client information (Sec.3.4).
The paper claims privacy preservation from prototype averaging without a formal privacy analysis.

how reviews work

0 comments

Cite this review

Pith. "Pith review of FedSC: Federated Learning with Semantic-Aware Collaboration." pith.science (2026). https://pith.science/paper/VIOXQVG4

@misc{pith2026250621012,
  author       = {Pith},
  title        = {Pith review of: FedSC: Federated Learning with Semantic-Aware Collaboration},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/VIOXQVG4}},
  note         = {Machine review of arXiv:2506.21012}
}

read the original abstract

Federated learning (FL) aims to train models collaboratively across clients without sharing data for privacy-preserving. However, one major challenge is the data heterogeneity issue, which refers to the biased labeling preferences at multiple clients. A number of existing FL methods attempt to tackle data heterogeneity locally (e.g., regularizing local models) or globally (e.g., fine-tuning global model), often neglecting inherent semantic information contained in each client. To explore the possibility of using intra-client semantically meaningful knowledge in handling data heterogeneity, in this paper, we propose Federated Learning with Semantic-Aware Collaboration (FedSC) to capture client-specific and class-relevant knowledge across heterogeneous clients. The core idea of FedSC is to construct relational prototypes and consistent prototypes at semantic-level, aiming to provide fruitful class underlying knowledge and stable convergence signals in a prototype-wise collaborative way. On the one hand, FedSC introduces an inter-contrastive learning strategy to bring instance-level embeddings closer to relational prototypes with the same semantics and away from distinct classes. On the other hand, FedSC devises consistent prototypes via a discrepancy aggregation manner, as a regularization penalty to constrain the optimization region of the local model. Moreover, a theoretical analysis for FedSC is provided to ensure a convergence guarantee. Experimental results on various challenging scenarios demonstrate the effectiveness of FedSC and the efficiency of crucial components.

Figures

Figures reproduced from arXiv: 2506.21012 by the authors.

**Figure 2.** An example of computing neighbor relationships [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗

**Figure 3.** Illustration of different class-wise prototypes. [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗

Figures from the paper (3 more)

**Figure 4.** Figure 4: The overview of Federated Learning with Semantic-Aware Collaboration (FedSC), including two complementary [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]

**Figure 5.** Figure 5: Effects of 𝜏 and 𝑀 on CIFAR10 dataset with NID10.2 [PITH_FULL_IMAGE:figures/full_fig_p009_5.png]

**Figure 6.** Figure 6: Ablation study on clients 𝐾 and local epochs 𝐸 [PITH_FULL_IMAGE:figures/full_fig_p009_6.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

77 extracted references · 70 canonical work pages

[1]

Kaidi Cao, Colin Wei, Adrien Gaidon, et al. 2019. Learning Imbalanced Datasets with Label-Distribution-Aware Margin Loss. InNeurIPS. 1567–1578. KDD ’25, August 3–7, 2025, Toronto, ON, Canada Huan Wang et al

work page 2019
[2]

Haokun Chen, Ahmed Frikha, Denis Krompass, Jindong Gu, and Volker Tresp

work page
[3]

Hong-You Chen and Wei-Lun Chao. 2021. FedBE: Making Bayesian Model Ensemble Applicable to Federated Learning. InICLR. https://openreview.net/ forum?id=dgtpE6gKjHn

work page 2021
[4]

Ting Chen, Simon Kornblith, Mohammad Norouzi, and Geoffrey Hinton. 2020. A simple framework for contrastive learning of visual representations. InICML. 1597–1607

2020
[5]

Yae Jee Cho, Jianyu Wang, and Gauri Joshi. 2022. Towards understanding biased client selection in federated learning. InAISTAT. 10351–10375

work page 2022
[6]

Yutong Dai, Zeyuan Chen, Junnan Li, Shelby Heinecke, Lichao Sun, and Ran Xu

work page
[7]

Pieter-Tjerk De Boer, Dirk P Kroese, Shie Mannor, and Reuven Y Rubinstein

work page
[8]

Tackling data heterogeneity in federated learning with class prototypes. In AAAI. 7314–7322

work page

Show all 77 references

[9]

Yann Fraboni, Richard Vidal, Laetitia Kameni, and Marco Lorenzi. 2021. Clustered sampling: Low-variance and improved representativity for clients selection in federated learning. InICML. 3407–3416

2021
[10]

Liang Gao, Huazhu Fu, Li Li, Yingwen Chen, Ming Xu, and Cheng-Zhong Xu
[11]

Jiankang Deng, Jia Guo, Jing Yang, et al. 2021. Variational prototype learning for deep face recognition. InCVPR. 11906–11915

2021
[12]

Kaiming He, Haoqi Fan, Yuxin Wu, Saining Xie, and Ross Girshick. 2020. Mo- mentum contrast for unsupervised visual representation learning. InCVPR. 9729– 9738

2020
[13]

Mingcheng Hou and Issei Sato. 2022. A closer look at prototype classifier for few-shot image classification.NeurIPS35 (2022), 25767–25778

2022
[14]

Cheng Ju, Aurélien Bibaut, and Mark van der Laan. 2018. The relative perfor- mance of ensemble methods with deep convolutional neural networks for image classification.Journal of Applied Statistics45 (2018), 2800–2818

2018
[15]

Weituo Hao, Mostafa El-Khamy, Jungwon Lee, Jianyi Zhang, et al. 2021. Towards fair federated learning with zero-shot data augmentation. InCVPR. 3310–3319

2021
[16]

Nazmul Karim, Mamshad Nayeem Rizve, Nazanin Rahnavard, Ajmal Mian, and Mubarak Shah. 2022. Unicon: Combating label noise through uniform selection and contrastive learning. InCVPR. 9676–9686

2022
[17]

Sai Praneeth Karimireddy, Satyen Kale, Mehryar Mohri, Sashank Reddi, Sebastian Stich, and Ananda Theertha Suresh. 2020. SCAFFOLD: Stochastic Controlled Averaging for Federated Learning. InICML. 5132–5143

2020
[18]

Alex Krizhevsky and Geoffrey Hinton. 2009. Learning Multiple Layers of Features from Tiny Images. InTechnical report, University of Toronto. Toronto, Ontario

2009
[19]

Peter Kairouz, H Brendan McMahan, Brendan Avent, Aurélien Bellet, Mehdi Bennis, Arjun Bhagoji, et al. 2021. Advances and open problems in federated learning.Found. Trends Mach. Learn.14, 1–2 (2021), 1–210

2021
[20]

Phuc H Le-Khac, Graham Healy, and Alan F Smeaton. 2020. Contrastive represen- tation learning: A framework and review.IEEE Access8 (2020), 193907–193934

2020
[21]

Gihun Lee, Minchan Jeong, Yongjin Shin, Sangmin Bae, and Se-Young Yun. 2022. Preservation of the global knowledge by not-true distillation in federated learning. NeurIPS35 (2022), 38461–38474

2022
[22]

Ang Li, Jingwei Sun, Xiao Zeng, Mi Zhang, Hai Li, and Yiran Chen. 2021. Fed- Mask: Joint Computation and Communication-Efficient Personalized Federated Learning via Heterogeneous Masking. InACM SenSys. 42–55

2021
[23]

Ya Le and Xuan Yang. 2015. Tiny imagenet visual recognition challenge.CS 231N 7, 7 (2015), 3

2015
[24]

Hao Li, Zheng Xu, Gavin Taylor, Christoph Studer, and Tom Goldstein. 2018. Visualizing the Loss Landscape of Neural Nets. InNeurIPS. 6391–6401

2018
[25]

Qinbin Li, Yiqun Diao, Quan Chen, and Bingsheng He. 2022. Federated Learning on Non-IID Data Silos: An Experimental Study. InICDE. 965–978

2022
[26]

Qinbin Li, Bingsheng He, and Dawn Song. 2021. Model-contrastive federated learning. InCVPR. 10713–10722

2021
[27]

Gen Li, Varun Jampani, Laura Sevilla-Lara, Deqing Sun, Jonghyun Kim, and Joongkyu Kim. 2021. Adaptive prototype learning and allocation for few-shot segmentation. InCVPR. 8334–8343

2021
[28]

Tian Li, Anit Kumar Sahu, Manzil Zaheer, Maziar Sanjabi, Ameet Talwalkar, and Virginia Smith. 2020. Federated optimization in heterogeneous networks.MLSys 2 (2020), 429–450

2020
[29]

Xiang Li, Kaixuan Huang, Wenhao Yang, Shusen Wang, and Zhihua Zhang. 2020. On the Convergence of FedAvg on Non-IID Data. InICLR. https://openreview. net/forum?id=HJxNAnVtDS

2020
[30]

Yunfan Li, Peng Hu, Zitao Liu, Dezhong Peng, Joey Tianyi Zhou, and Xi Peng
[31]

Tian Li, Anit Kumar Sahu, Ameet Talwalkar, and Virginia Smith. 2020. Federated learning: Challenges, methods, and future directions.IEEE Signal Processing Magazine37, 3 (2020), 50–60

2020
[32]

Tao Lin, Lingjing Kong, Sebastian U Stich, et al. 2020. Ensemble distillation for robust model fusion in federated learning.NeurIPS33 (2020), 2351–2363

2020
[33]

Brendan McMahan, Eider Moore, Daniel Ramage, Seth Hampson, et al . 2017. Communication-Efficient Learning of Deep Networks from Decentralized Data. InAISTAT. 1273–1282

2017
[34]

Rui Miao, Yintao Yang, Yao Ma, Xin Juan, Haotian Xue, Jiliang Tang, Ying Wang, and Xin Wang. 2022. Negative samples selecting strategy for graph contrastive learning.Information Sciences613 (2022), 667–681

2022
[35]

Xutong Mu, Yulong Shen, Ke Cheng, Xueli Geng, Jiaxuan Fu, Tao Zhang, and Zhiwei Zhang. 2023. Fedproc: Prototypical contrastive federated learning on non-iid data.Future Generation Computer Systems143 (2023), 93–104

2023
[36]

Chengzhi Lin, Ancong Wu, Junwei Liang, Jun Zhang, Wenhang Ge, Wei-Shi Zheng, and Chunhua Shen. 2022. Text-adaptive multiple visual prototype match- ing for video-text retrieval.NeurIPS35 (2022), 38655–38666

2022
[37]

John Nguyen, Kshitiz Malik, Hongyuan Zhan, Ashkan Yousefpour, Mike Rab- bat, Mani Malek, and Dzmitry Huba. 2022. Federated learning with buffered asynchronous aggregation. InAISTAT. 3581–3607

2022
[38]

Aaron van den Oord, Yazhe Li, and Oriol Vinyals. 2018. Representation learning with contrastive predictive coding.arXiv preprint arXiv:1807.03748(2018)

2018 arXiv
[39]

Boris Oreshkin, Pau Rodríguez López, and Alexandre Lacoste. 2018. Tadam: Task dependent adaptive metric for improved few-shot learning.NeurIPS31 (2018), 719–729

2018
[40]

Mark Sandler, Andrew Howard, Menglong Zhu, Andrey Zhmoginov, and Liang- Chieh Chen. 2018. Mobilenetv2: Inverted residuals and linear bottlenecks. In CVPR. 4510–4520

2018
[41]

Nguyen, Chien Duong, Tu Phan, Erman Tjiputra, and Quang D

Anh Nguyen, Tuong Do, Minh Tran, Binh X. Nguyen, Chien Duong, Tu Phan, Erman Tjiputra, and Quang D. Tran. 2022. Deep Federated Learning for Au- tonomous Driving. InIEEE Intelligent Vehicles Symposium. 1824–1830

2022
[42]

Jake Snell, Kevin Swersky, and Richard Zemel. 2017. Prototypical Networks for Few-shot Learning. InNeurIPS. 4080–4090

2017
[43]

Yue Tan, Yixin Liu, Guodong Long, Jing Jiang, Qinghua Lu, and Chengqi Zhang
[44]

Yue Tan, Guodong Long, Lu Liu, Tianyi Zhou, Qinghua Lu, Jing Jiang, and Chengqi Zhang. 2022. Fedproto: Federated prototype learning across heteroge- neous clients. InAAAI. 8432–8440

2022
[45]

Yue Tan, Guodong Long, Jie Ma, Lu Liu, Tianyi Zhou, and Jing Jiang. 2022. Federated learning from pre-trained models: A contrastive learning approach. NeurIPS35 (2022), 19332–19344

2022
[46]

Seonguk Seo, Jinkyu Kim, Geeho Kim, and Bohyung Han. 2024. Relaxed con- trastive learning for federated learning. InCVPR. 12279–12288

2024
[47]

Haozhao Wang, Haoran Xu, Yichen Li, Yuan Xu, Ruixuan Li, and Tianwei Zhang
[48]

Hongyi Wang, Mikhail Yurochkin, Yuekai Sun, Dimitris Papailiopoulos, and Yasaman Khazaeni. 2020. Federated Learning with Matched Averaging. InICLR. https://openreview.net/forum?id=BkluqlSFDS

2020
[49]

Federated learning on non-iid graphs via structural knowledge sharing. In AAAI. 9953–9961
[50]

Yujie Wei, Jiaxin Ye, Zhizhong Huang, Junping Zhang, and Hongming Shan. 2023. Online prototype learning for online continual learning. InICCV. 18764–18774

2023
[51]

Jiazheng Xing, Mengmeng Wang, Yudi Ruan, Bofan Chen, Yaowei Guo, Boyu Mu, Guang Dai, Jingdong Wang, and Yong Liu. 2023. Boosting few-shot action recognition with graph-guided hybrid matching. InICCV. 1740–1750

2023
[52]

Huan Wang, Lijuan Wang, and Jun Shen. 2022. Logit Calibration for Non-IID and Long-Tailed Data in Federated Learning. InIEEE ISPA. 782–789

2022
[53]

Xilie Xu, Jingfeng Zhang, Feng Liu, Masashi Sugiyama, and Mohan S Kankanhalli
[54]

Wanqi Xue and Wei Wang. 2020. One-shot image classification by learning to restore prototypes. InAAAI. 6558–6565

2020
[55]

Hong-Ming Yang, Xu-Yao Zhang, Fei Yin, and Cheng-Lin Liu. 2018. Robust classification with convolutional prototype learning. InCVPR. 3474–3482

2018
[56]

Xiaohan Wang, Linchao Zhu, Heng Wang, and Yi Yang. 2021. Interactive proto- type learning for egocentric action recognition. InICCV. 8168–8177

2021
[57]

Chun-Han Yao, Boqing Gong, Hang Qi, Yin Cui, Yukun Zhu, and Ming-Hsuan Yang. 2022. Federated multi-target domain adaptation. InW ACV. 1424–1433

2022
[58]

Rui Ye, Mingkai Xu, Jianyu Wang, Chenxin Xu, Siheng Chen, and Yanfeng Wang
[59]

Wenjia Xu, Yongqin Xian, Jiuniu Wang, et al. 2020. Attribute prototype network for zero-shot learning.NeurIPS33 (2020), 21969–21980

2020
[60]

Xiaotong Yuan and Ping Li. 2022. On convergence of FedProx: Local dissimilarity invariant bounds, non-smoothness and beyond.NeurIPS35 (2022), 10752–10765. FedSC: Federated Learning with Semantic-Aware Collaboration KDD ’25, August 3–7, 2025, Toronto, ON, Canada

2022
[61]

Efficient adversarial contrastive learning via robustness-aware coreset selection.NeurIPS36 (2024), 75798–75825

2024
[62]

Han Zhao, Xu Yang, Zhenru Wang, Erkun Yang, and Cheng Deng. 2021. Graph Debiased Contrastive Learning with Joint Representation Clustering. InIJCAI. 3434–3440

2021
[63]

Kecheng Zheng, Cuiling Lan, Wenjun Zeng, Jiawei Liu, Zhizheng Zhang, and Zheng-Jun Zha. 2021. Pose-guided feature learning with knowledge distillation for occluded person re-identification. InACM MM. 4537–4545

2021
[64]

Xihong Yang, Yue Liu, Sihang Zhou, Siwei Wang, Wenxuan Tu, Qun Zheng, Xinwang Liu, Liming Fang, and En Zhu. 2023. Cluster-guided contrastive graph clustering network. InAAAI. 10834–10842

2023
[65]

Fei Zhu, Xu-Yao Zhang, Chuang Wang, Fei Yin, and Cheng-Lin Liu. 2021. Pro- totype augmentation and self-supervision for incremental learning. InCVPR. 5871–5880

2021
[66]

Hangyu Zhu, Jinjin Xu, Shiqing Liu, and Yaochu Jin. 2021. Federated learning on non-IID data: A survey.Neurocomputing465 (2021), 371–390

2021
[67]

FedDisco: Federated learning with discrepancy-aware collaboration. In ICML. 39879–39902
[68]

Yang Yu, Qi Liu, Likang Wu, Runlong Yu, Sanshi Lei Yu, and Zaixi Zhang. 2023. Untargeted attack against federated recommendation systems via poisonous item embeddings and the defense. InAAAI. 4854–4863

2023
[70]

Lin Zhang, Li Shen, Liang Ding, Dacheng Tao, and Ling-Yu Duan. 2022. Fine- tuning global model via data-free knowledge distillation for non-iid federated learning. InCVPR. 10174–10183

2022
[73]

Tianfei Zhou and Wenguan Wang. 2024. Prototype-based semantic segmentation. IEEE TPAMI(2024), 6858–6872

2024
[76]

Zhuangdi Zhu, Junyuan Hong, and Jiayu Zhou. 2021. Data-free knowledge distillation for heterogeneous federated learning. InICML. 12878–12889

2021
[77]

Fuzhen Zhuang, Zhiyuan Qi, Keyu Duan, Dongbo Xi, Yongchun Zhu, et al. 2020. A comprehensive survey on transfer learning.Proc. IEEE109 (2020), 43–76. A Theoretical Analysis A.1 Preliminaries We expresse ∈ {1, 2,...,𝐸} as the local iteration step, 𝑟 as the global round,𝐸 as the ...

2020
[2005]

A tutorial on the cross-entropy method.Annals of Operations Research134 (2005), 19–67

2005
[2021]

Contrastive clustering. InAAAI. 8547–8555
[2022]

FedDC: Federated Learning with Non-IID Data via Local Drift Decoupling and Correction. InCVPR. 10102–10111
[2023]

FRAug: Tackling federated learning with Non-IID features via representa- tion augmentation. InICCV. 4849–4859
[2024]

FedCDA: Federated Learning with Cross-rounds Divergence-aware Aggre- gation. InICLR. https://openreview.net/forum?id=nbPGqeH3lt

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.

[1] [1]

Kaidi Cao, Colin Wei, Adrien Gaidon, et al. 2019. Learning Imbalanced Datasets with Label-Distribution-Aware Margin Loss. InNeurIPS. 1567–1578. KDD ’25, August 3–7, 2025, Toronto, ON, Canada Huan Wang et al

work page 2019

[2] [2]

Haokun Chen, Ahmed Frikha, Denis Krompass, Jindong Gu, and Volker Tresp

work page

[3] [3]

Hong-You Chen and Wei-Lun Chao. 2021. FedBE: Making Bayesian Model Ensemble Applicable to Federated Learning. InICLR. https://openreview.net/ forum?id=dgtpE6gKjHn

work page 2021

[4] [4]

Ting Chen, Simon Kornblith, Mohammad Norouzi, and Geoffrey Hinton. 2020. A simple framework for contrastive learning of visual representations. InICML. 1597–1607

2020

[5] [5]

Yae Jee Cho, Jianyu Wang, and Gauri Joshi. 2022. Towards understanding biased client selection in federated learning. InAISTAT. 10351–10375

work page 2022

[6] [6]

Yutong Dai, Zeyuan Chen, Junnan Li, Shelby Heinecke, Lichao Sun, and Ran Xu

work page

[7] [7]

Pieter-Tjerk De Boer, Dirk P Kroese, Shie Mannor, and Reuven Y Rubinstein

work page

[8] [8]

Tackling data heterogeneity in federated learning with class prototypes. In AAAI. 7314–7322

work page