Pith. sign in

REVIEW 4 major objections 5 minor 43 references

EFTViT: Efficient Federated Training of Vision Transformers with Masked Images on Resource-Constrained Clients

T0 review · 4 major / 5 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read Federated ViT training can be made 5.6x cheaper on clients by masking 75% of each image, while improving accuracy by up to 2.46% over prior resource-constrained methods.

desk verdict Useful combination of known components that works in experiments, but the complexity analysis in §3.6 is wrong; the actual savings come from masking, not from freezing the global module. read the letter →

arxiv 2412.00334 v2 pith:PEHUSVJN submitted 2024-11-30 cs.CV cs.AI

Meihan Wu , Tao Chang , Cui Miao , Jie Zhou , Chun Li , Xiangyu Xu , Ming Li , Xiaodong Wang This is my paper · ORCID
classification cs.CVcs.AI
keywords federatedlearningvisiontransformermaskedimagemodelingresource-constrainedclientshierarchicaltrainingdataheterogeneityintermediatefeatureprivacyparameter-efficientfine-tuning
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

EFTViT is a proposal for training Vision Transformers in federated settings where client devices are too weak to run the full model. The paper argues that images are redundant enough that randomly masking 75% of the patches before local training barely hurts accuracy while cutting local computation by up to 5.6x and local training time by up to 3.1x. To make this work, the model is split into lightweight local modules on clients and a larger global module on the server, with the server trained on intermediate patch features rebalanced by a median sampling strategy. If the claims hold, resource-constrained edge clients could train all ViT parameters collaboratively rather than falling back to parameter-efficient fine-tuning, and the masked patches also make uploaded features harder to reconstruct into images. The experiments report accuracy gains up to 2.46% over existing resource-constrained FL methods on CIFAR-10, CIFAR-100, and UC Merced Land-Use.

What carries the argument

The central object is the masking-plus-hierarchy training scheme. Images are patchified into 16x16 tokens, a fraction $r_m = 0.75$ of patches is randomly dropped, and only the remaining tokens, with positional embeddings preserved, pass through the client's local module of $M = 2$ Transformer layers; the server's global module of $N = 10$ layers is trained on the uploaded patch features. A median sampling strategy balances the patch-feature dataset so each class contributes the same number of samples, hiding client data-distribution statistics. The complexity analysis ties the savings to the formula $\mathcal{O}((15N_T - 10N)(1 - r_m) n d^2 + (6N_T - 4N)(1 - r_m)^2 n^2 d)$, where $N_T$ is the total number of Transformer layers, $n$ the number of patches, and $d$ the latent dimension, so both a larger server-side share $N$ and a higher masking ratio $r_m$ reduce client cost.

What would settle it

Run EFTViT with $r_m = 0.75$ on a dataset whose discriminative information is spread uniformly across the whole image (e.g., fine-grained texture or medical pathology images) and compare against Fed-Full: if the accuracy gap grows well beyond the roughly 1-2% seen on CIFAR and UC Merced, the redundancy premise is falsified. A cheaper check is to sweep the masking ratio on any new dataset and see whether accuracy collapses before 75% is reached.

Watch

Extended reading notes

Core claim

The central claim is that ViT training for federated classification can be made cheap enough for resource-constrained clients by exploiting image redundancy: divide each image into non-overlapping patches, randomly discard 75% of them, and train the client-side layers only on the remaining quarter. The paper reports that this masking step reduces the client's forward-pass cost roughly fivefold with minimal accuracy loss, and that the saved budget can be reinvested in training all model parameters across the hierarchy instead of only a small tunable subset. Concretely, each client trains two Transformer layers and a classification head on masked patches, while the server trains the remaining ten Transformer layers on patch features that clients balance across classes using the median class count before upload. Across CIFAR-10, CIFAR-100, and UC Merced Land-Use with high and low data heterogeneity, EFTViT is reported to beat parameter-efficient baselines (head-only, bias, prompt, LoRA) and partial-training methods (FEDBFPT, FedRA), reaching target accuracy in fewer rounds and with 5.6x lower local computation and 3.1x lower per-round training time.

Load-bearing premise

The load-bearing premise is that images carry enough redundancy that randomly discarding 75% of the patches before training barely hurts classification accuracy, and if that fails on a new dataset, at a different patch size, or under stronger data heterogeneity, the efficiency gains will come at an unacceptable accuracy cost.

Editorial extensions

If this is right

  • Client devices can participate in full-parameter ViT training instead of being limited to small tunable subsets, which is what the accuracy gains over head, bias, prompt, and LoRA baselines are attributed to.
  • Uploaded 25%-patch features are harder to turn back into images, so masking acts as a built-in content-privacy layer on top of the usual intermediate-feature exchange.
  • Median sampling equalizes class counts before upload, so the server cannot infer client preference skew from data distribution statistics.
  • Convergence to 85% accuracy is claimed in as few as 3 to 6 rounds on the tested benchmarks versus 9 to 119 rounds for baselines, which translates to large reductions in total wall-clock time including communication.
  • The method's accuracy changes little when the local module depth $M$ is 2, 4, or 6 layers, so the same framework adapts to clients with different compute budgets.

Reading between the lines

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

  • Beyond the three tested benchmarks, the 75% masking sweet spot should transfer only to datasets whose classification signal is spatially redundant; fine-grained or small-object datasets are the natural stress test.
  • The reconstruction-attack experiment is heuristic, so the privacy claim is best read as 'harder to reconstruct' rather than a formal guarantee; a determined attacker with a learned image prior might still recover coarse content.
  • The efficiency comparison is against PEFT and partial-training baselines, not against full-parameter FedAvg; reading the reported numbers against the Fed-Full row suggests the honest trade is roughly 0.3 to 2.3% accuracy for a 4x client-compute reduction.
  • Median sampling hides per-class counts, but it may still leak correlations between features and labels or class co-occurrence; differential privacy, which the paper notes is orthogonal, could close that gap.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper proposes EFTViT, a hierarchical federated learning framework for Vision Transformers on resource-constrained clients. Each client trains a lightweight M-layer local module and a classification head on randomly masked image patches (default rm=0.75), while a central server trains the N-layer global module on median-sampled intermediate patch features uploaded from clients. The authors claim that EFTViT reduces local training computational cost by up to 5.6x, reduces local training time by up to 3.1x, improves accuracy by up to 2.46% over existing resource-constrained FL baselines, and provides privacy benefits through masking and the median sampling strategy. Experiments on CIFAR-10, CIFAR-100, and UC Merced Land-Use under two Dirichlet heterogeneity settings are reported in Tables 1-3 and Figures 4-7.

Significance. If the reported results hold, EFTViT is a practically valuable recipe for federated ViT training on edge clients: it combines an image-masking prior with a hierarchical parameter split and a class-balancing upload mechanism, and it shows consistent gains over several PEFT-based baselines on three benchmarks. The paper's strengths are its clear motivation (Figure 2), the breadth of compared methods, the convergence-speed analysis, and the fact that the central mechanism is simple and easy to implement. However, the absence of multiple seeds, the lack of code, the internal inconsistency in the complexity accounting, and the unresolved abstract discrepancy mean that the headline efficiency and accuracy claims should be treated with caution.

major comments (4)
  1. [Section 3.6, Eq. (1)-(2)] The backward-propagation complexity in Section 3.6 is undercounted. Because the client loss in Eq. (1) is computed after the frozen N-layer global module, gradients for the trainable local-module parameters phi_k must be backpropagated through all NT layers, even though no parameter gradients are accumulated for the frozen layers. The backward term O(10*(NT-N)*(1-rm)*n*d^2 + ...) should therefore scale with NT, not NT-N; with NT=12 and N=10 this undercounts backward work by roughly a factor of 6. The reported GFLOPs in Table 3 (2.997, close to 12.005/4) appear to correspond to a forward-only pass, so the paper should state explicitly that the GFLOPs column is forward-only and either correct the complexity formula or qualify the abstract's 'computational cost' claim to match the measured TTPR.
  2. [Abstract and Section 1] The listing abstract reports 'up to 28.17% accuracy improvement, reduces local training computational cost by up to 2.8x, and cuts local training time by up to 4.4x', whereas the full-text abstract and Section 1 report 'reduces the computational cost of local training by as much as 5.6x, lowers the local training time by up to 3.1x, and improves the accuracy by up to 2.46%'. These are materially different claims and must be reconciled; as presented, a reader cannot tell which numbers are the actual results.
  3. [Section 4.3, Tables 1 and 3] All experimental numbers appear to come from single runs with no error bars or significance tests. Several decisive comparisons are very small: on CIFAR-10 with beta=0.1, EFTViT (98.12) is only 0.13 points above Fed-Prompt and FEDBFPT (97.99), and on CIFAR-100 with beta=1.0 the margin over Fed-Bias is 0.14 points. In addition, the key hyperparameters rm, M, and the sampling threshold are selected by ablations on the same test benchmarks (Figures 6-7, Tables 4-5), which makes the reported accuracies optimistic. The 'state-of-the-art' and 'consistently outperforms' claims need multi-seed results to be convincing.
  4. [Section 4.3, Privacy Analysis] The privacy evaluation is qualitative only: it shows reconstructed images and states that they are 'severely distorted' but provides no quantitative metric (e.g., PSNR/SSIM) and no comparison against reconstruction from full features or from gradients. Also, the claim that median sampling 'erases' data distribution privacy is stronger than what is demonstrated: the strategy equalizes per-class sample counts, but other distributional information in the uploaded features is not analyzed.
minor comments (5)
  1. [Eq. (1)] The loss in Eq. (1) is written as a sum over classes p(y=i) log(...) but does not index over samples or divide by the batch size; please rewrite it as an empirical expectation over the local data.
  2. [Section 3.3] The statement that 'the entire images are adopted for inference on each client' while training uses only 25% of patches creates a train/inference distribution shift that should be discussed explicitly, since it affects how the reported accuracy should be interpreted.
  3. [Figure 2] The caption says masking to 0.75 'minimally affects' performance, but the reported CIFAR-100 drop is 3.41 points; please qualify this as a small but non-negligible degradation.
  4. [Section 4.2] No code or repository is provided, and the paper does not state whether the reported numbers are averaged over multiple runs; please add reproducibility details, including the number of seeds and hardware/software versions.
  5. [Table 3 and Section 3.6] The distinction between the GFLOPs column (forward propagation only, per Section 4.3) and the complexity analysis in Section 3.6 (forward plus backward) should be made explicit in the table caption and in the abstract's use of the term 'computational cost'.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity found: EFTViT's claims rest on empirical masked-image motivation, measured baselines, and arithmetic FLOPs, not on a self-referential derivation.

full rationale

EFTViT's derivation chain is empirical and compositional, not self-referential. The masked-image premise is established by the unconstrained FL experiments in Fig. 2 and by external masked-image-modeling results (MAE [10], [31]); the final efficiency numbers are either standard GFLOPs arithmetic from the stated masking ratio and baseline configurations or measured client runtimes compared against external methods (FedRA, FEDBFPT, and PEFT baselines). No fitted parameter is renamed as a prediction: the masking ratio, local module depth, and median sampling threshold are chosen through ablations on the same benchmarks, which creates selection bias but does not make the reported accuracies equivalent to the chosen hyperparameters by construction. The only self-citation, FedEKT [30], is used to support the claim that intermediate-feature exchange is safer than gradient exchange, but that claim is jointly attributed to the external FedGKT [8], so the self-citation is not load-bearing. The Section 3.6 complexity formula is internally questionable because it counts backward cost only for the (NT-N) trainable local layers, while Eq. (1) computes the loss through all NT layers including the frozen global module; this is a correctness/fairness risk for the 5.6x savings claim, not a circularity, since the formula is not derived from the claimed savings. Overall, no load-bearing step reduces to its own inputs.

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

All numerical results rest on a pretrained ViT-B, Dirichlet partitions, and hyperparameters selected on the evaluation benchmarks. No new physical or architectural entities are postulated.

free parameters (4)
  • masking ratio r_m = 0.75
    Selected from ablation on CIFAR-100 and UC Merced (Fig. 6) to balance compute reduction against accuracy drop.
  • local module depth M = 2
    Chosen from Table 5; M=2 gives the lowest client parameters with near-best accuracy.
  • global module depth N = 10
    Default configuration paired with M=2 for a 12-layer ViT-B; not ablated independently.
  • median sampling threshold = median class count
    Selected from Fig. 7; higher thresholds add server compute without meaningful accuracy gain.
assumptions (4)
  • ad hoc to paper Images contain sufficient redundancy that training on 25% of patches preserves recognition accuracy.
    The central efficiency premise; tested in Fig. 2 on two benchmarks without resource constraints before being adopted as the default.
  • domain assumption Intermediate patch features are a safe substrate for server training, and sharing them is less risky than sharing gradients.
    Privacy analysis in Section 3.6 relies on citations [8,30] and a reconstruction simulation, not a formal guarantee.
  • domain assumption ViT-B pretrained on ImageNet-21K provides a strong initialization for all federated clients.
    All methods use the same pretrained backbone, and reported accuracies depend on this starting point.
  • domain assumption Dirichlet partitions with beta in {0.1, 1} model realistic data heterogeneity.
    Standard FL benchmark setup; results may not extend to other heterogeneity patterns or label skews.

how reviews work

0 comments
Cite this review

Pith. "Pith review of EFTViT: Efficient Federated Training of Vision Transformers with Masked Images on Resource-Constrained Clients." pith.science (2026). https://pith.science/paper/PEHUSVJN

@misc{pith2026241200334,
  author       = {Pith},
  title        = {Pith review of: EFTViT: Efficient Federated Training of Vision Transformers with Masked Images on Resource-Constrained Clients},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/PEHUSVJN}},
  note         = {Machine review of arXiv:2412.00334}
}
abstract

Federated learning research has recently shifted from Convolutional Neural Networks (CNNs) to Vision Transformers (ViTs) due to their superior capacity. ViTs training demands higher computational resources due to the lack of 2D inductive biases inherent in CNNs. However, efficient federated training of ViTs on resource-constrained edge devices remains unexplored in the community. In this paper, we propose EFTViT, a hierarchical federated framework that leverages masked images to enable efficient, full-parameter training on resource-constrained edge devices, offering substantial benefits for learning on heterogeneous data. In general, we patchify images and randomly mask a portion of the patches, observing that excluding them from training has minimal impact on performance while substantially reducing computation costs and enhancing data content privacy protection. Specifically, EFTViT comprises a series of lightweight local modules and a larger global module, updated independently on clients and the central server, respectively. The local modules are trained on masked image patches, while the global module is trained on intermediate patch features uploaded from the local client, balanced through a proposed median sampling strategy to erase client data distribution privacy. We analyze the computational complexity and privacy protection of EFTViT. Extensive experiments on popular benchmarks show that EFTViT achieves up to 28.17% accuracy improvement, reduces local training computational cost by up to 2.8$\times$, and cuts local training time by up to 4.4$\times$ compared to existing methods.

Figures

Figures reproduced from arXiv: 2412.00334 by the authors.

Figure 2
Figure 2. Impact of varying masking ratio rm in federated learn￾ing experiments conducted on two benchmarks without resource constraints. Experiments are conducted under a high data hetero￾geneity setting with β = 0.1, the concentration parameter of the Dirichlet distribution DirN (β). Results indicate that increasing the masking ratio to 0.75 minimally affects model performance while substantially reducing training costs. In… view at source ↗
Figure 3
Figure 3. Overview of our hierarchical framework, EFTViT, for efficient federated training of ViTs on resource-constrained clients. Local [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figure 4
Figure 4. Testing accuracy progression of EFTViT and other baselines under high data heterogeneity ( [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
Figures from the paper (3 more)
Figure 5
Figure 5. Figure 5: Visualization of reconstructed images from patch fea [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 7
Figure 7. Figure 7: Accuracy of our approach with different sampling [PITH_FULL_IMAGE:figures/full_fig_p008_7.png]
Figure 6
Figure 6. Figure 6: Accuracy changes of EFTViT with varying masking ra [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

43 extracted references · 27 canonical work pages

  1. [1]

    Fe- drolex: Model-heterogeneous federated learning with rolling sub-model extraction

    Samiul Alam, Luyang Liu, Ming Yan, and Mi Zhang. Fe- drolex: Model-heterogeneous federated learning with rolling sub-model extraction. Advances in Neural Information Pro- cessing Systems, 35:29677–29690, 2022. 2

    work page 2022

  2. [2]

    Dearkd: data-efficient early knowledge distillation for vision transformers

    Xianing Chen, Qiong Cao, Yujie Zhong, Jing Zhang, Shenghua Gao, and Dacheng Tao. Dearkd: data-efficient early knowledge distillation for vision transformers. In Pro- ceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 12052–12062, 2022. 1

    work page 2022

  3. [3]

    Up-detr: Unsupervised pre-training for object detection with transformers

    Zhigang Dai, Bolun Cai, Yugeng Lin, and Junying Chen. Up-detr: Unsupervised pre-training for object detection with transformers. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages 1601– 1610, 2021. 1

    work page 2021

  4. [4]

    Heterofl: Com- putation and communication efficient federated learning for heterogeneous clients

    Enmao Diao, Jie Ding, and Vahid Tarokh. Heterofl: Com- putation and communication efficient federated learning for heterogeneous clients. In International Conference on Learning Representations, 2020. 2

    work page 2020

  5. [5]

    An image is worth 16x16 words: Trans- formers for image recognition at scale

    Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Syl- vain Gelly, et al. An image is worth 16x16 words: Trans- formers for image recognition at scale. arXiv preprint arXiv:2010.11929, 2020. 1, 5

    arXiv 2010

  6. [6]

    Differential privacy: A survey of results

    Cynthia Dwork. Differential privacy: A survey of results. In International conference on theory and applications of mod- els of computation, pages 1–19. Springer, 2008. 5

    work page 2008

  7. [7]

    Parameter-efficient fine-tuning for large models: A compre- hensive survey

    Zeyu Han, Chao Gao, Jinyang Liu, Sai Qian Zhang, et al. Parameter-efficient fine-tuning for large models: A compre- hensive survey. arXiv preprint arXiv:2403.14608, 2024. 3

    arXiv 2024

  8. [8]

    Group knowledge transfer: Federated learning of large cnns at the edge

    Chaoyang He, Murali Annavaram, and Salman Avestimehr. Group knowledge transfer: Federated learning of large cnns at the edge. Advances in Neural Information Processing Sys- tems, 33:14068–14080, 2020. 2, 3, 4

    work page 2020

Show all 43 references
  1. [9]

    Deep residual learning for image recognition

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In Proceed- ings of the IEEE conference on computer vision and pattern recognition, pages 770–778, 2016. 1

    2016

  2. [10]

    Masked autoencoders are scalable vision learners

    Kaiming He, Xinlei Chen, Saining Xie, Yanghao Li, Piotr Doll´ar, and Ross Girshick. Masked autoencoders are scalable vision learners. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages 16000– 16009, 2022. 1, 2

    2022

  3. [11]

    Lora: Low-rank adaptation of large language models

    Edward J Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen- Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen. Lora: Low-rank adaptation of large language models. arXiv preprint arXiv:2106.09685, 2021. 3, 5, 6, 7

    2021 arXiv

  4. [12]

    Adaptivefl: Adaptive heterogeneous federated learning for resource-constrained aiot systems

    Chentao Jia, Ming Hu, Zekai Chen, Yanxin Yang, Xiaofei Xie, Yang Liu, and Mingsong Chen. Adaptivefl: Adaptive heterogeneous federated learning for resource-constrained aiot systems. In Proceedings of the 61st ACM/IEEE Design Automation Conference, pages 1–6, 2024. 2

    2024

  5. [13]

    Vi- sual prompt tuning

    Menglin Jia, Luming Tang, Bor-Chun Chen, Claire Cardie, Serge Belongie, Bharath Hariharan, and Ser-Nam Lim. Vi- sual prompt tuning. In European Conference on Computer Vision, pages 709–727. Springer, 2022. 3, 5, 6, 7

    2022

  6. [14]

    Learning multiple layers of features from tiny images

    Alex Krizhevsky, Geoffrey Hinton, et al. Learning multiple layers of features from tiny images. 2009. 5

    2009

  7. [15]

    Unif2ace: Fine- grained face understanding and generation with unified mul- timodal models

    Junzhe Li, Xuerui Qiu, Linrui Xu, Liya Guo, Delin Qu, Tingting Long, Chun Fan, and Ming Li. Unif2ace: Fine- grained face understanding and generation with unified mul- timodal models. arXiv preprint arXiv:2503.08120, 2025. 3

    2025

  8. [16]

    Self- supervised geometric features discovery via interpretable at- tention for vehicle re-identification and beyond

    Ming Li, Xinming Huang, and Ziming Zhang. Self- supervised geometric features discovery via interpretable at- tention for vehicle re-identification and beyond. In ICCV,

  9. [17]

    Exploiting multi-view part-wise correlation via an ef- ficient transformer for vehicle re-identification

    Ming Li, Jun Liu, Ce Zheng, Xinming Huang, and Ziming Zhang. Exploiting multi-view part-wise correlation via an ef- ficient transformer for vehicle re-identification. TOM, 2021. 3

    2021

  10. [18]

    Dr-fer: Discrimina- tive and robust representation learning for facial expression recognition

    Ming Li, Huazhu Fu, Shengfeng He, Hehe Fan, Jun Liu, Jussi Keppo, and Mike Zheng Shou. Dr-fer: Discrimina- tive and robust representation learning for facial expression recognition. IEEE Transactions on Multimedia , 26:6297– 6309, 2023. 3

    2023

  11. [19]

    Stprivacy: Spatio-temporal privacy-preserving action recognition

    Ming Li, Xiangyu Xu, Hehe Fan, Pan Zhou, Jun Liu, Jia-Wei Liu, Jiahe Li, Jussi Keppo, Mike Zheng Shou, and Shuicheng Yan. Stprivacy: Spatio-temporal privacy-preserving action recognition. In ICCV, 2023. 3

    2023

  12. [20]

    Instant3d: instant text- to-3d generation

    Ming Li, Pan Zhou, Jia-Wei Liu, Jussi Keppo, Min Lin, Shuicheng Yan, and Xiangyu Xu. Instant3d: instant text- to-3d generation. IJCV, 2024. 3

    2024

  13. [21]

    Finch: Enhancing federated learning with hierarchical neural architecture search

    Jianchun Liu, Jiaming Yan, Hongli Xu, Zhiyuan Wang, Jinyang Huang, and Yang Xu. Finch: Enhancing federated learning with hierarchical neural architecture search. IEEE Transactions on Mobile Computing, 23(5):6012–6026, 2023. 2

    2023

  14. [22]

    Eventgpt: Event stream understanding with multimodal large language models

    Shaoyu Liu, Jianing Li, Guanghui Zhao, Yunjian Zhang, Xin Meng, Fei Richard Yu, Xiangyang Ji, and Ming Li. Eventgpt: Event stream understanding with multimodal large language models. In Proceedings of the Computer Vision and Pattern Recognition Conference (CVPR), pages 29139–2...

    2025

  15. [23]

    Realera: Semantic-level concept erasure via neighbor-concept min- ing

    Yufan Liu, Jinyang An, Wanqian Zhang, Ming Li, Dayan Wu, Jingzi Gu, Zheng Lin, and Weiping Wang. Realera: Semantic-level concept erasure via neighbor-concept min- ing. arXiv preprint arXiv:2410.09140, 2024. 3

    2024 arXiv

  16. [24]

    Communication- efficient learning of deep networks from decentralized data

    Brendan McMahan, Eider Moore, Daniel Ramage, Seth Hampson, and Blaise Aguera y Arcas. Communication- efficient learning of deep networks from decentralized data. In Artificial intelligence and statistics , pages 1273–1282. PMLR, 2017. 1, 6

    2017

  17. [25]

    Imagenet-21k pretraining for the masses

    Tal Ridnik, Emanuel Ben-Baruch, Asaf Noy, and Lihi Zelnik-Manor. Imagenet-21k pretraining for the masses. arXiv preprint arXiv:2104.10972, 2021. 5

    2021 arXiv

  18. [26]

    How to train your vit? data, augmentation, and regularization in vision transformers

    Andreas Steiner, Alexander Kolesnikov, Xiaohua Zhai, Ross Wightman, Jakob Uszkoreit, and Lucas Beyer. How to train your vit? data, augmentation, and regularization in vision transformers. arXiv preprint arXiv:2106.10270, 2021. 4

    2021 arXiv

  19. [27]

    Fedra: A ran- dom allocation strategy for federated tuning to unleash the power of heterogeneous clients

    Shangchao Su, Bin Li, and Xiangyang Xue. Fedra: A ran- dom allocation strategy for federated tuning to unleash the power of heterogeneous clients. In European Conference on Computer Vision, pages 342–358. Springer, 2024. 5, 6, 7

    2024

  20. [28]

    Conquering the communication constraints to enable large pre-trained models in federated learning.arXiv preprint arXiv:2210.01708, 2022

    Guangyu Sun, Matias Mendieta, Taojiannan Yang, and Chen Chen. Conquering the communication constraints to enable large pre-trained models in federated learning.arXiv preprint arXiv:2210.01708, 2022. 3

    2022 arXiv

  21. [29]

    Training data-efficient image transformers & distillation through at- tention

    Hugo Touvron, Matthieu Cord, Matthijs Douze, Francisco Massa, Alexandre Sablayrolles, and Herv ´e J´egou. Training data-efficient image transformers & distillation through at- tention. In International conference on machine learning , pages 10347–10357. PMLR, 2021. 1

    2021

  22. [30]

    Fedekt: Ensem- ble knowledge transfer for model-heterogeneous federated learning

    Meihan Wu, Li Li, Tao Chang, Peng Qiao, Cui Miao, Jie Zhou, Jingnan Wang, and Xiaodong Wang. Fedekt: Ensem- ble knowledge transfer for model-heterogeneous federated learning. In 2024 IEEE/ACM 32nd International Symposium on Quality of Service (IWQoS), pages 1–10. IEEE, 2024. 2, 4

    2024

  23. [31]

    Revealing the dark secrets of masked im- age modeling

    Zhenda Xie, Zigang Geng, Jingcheng Hu, Zheng Zhang, Han Hu, and Yue Cao. Revealing the dark secrets of masked im- age modeling. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages 14475– 14485, 2023. 2

    2023

  24. [32]

    Fedbfpt: An efficient federated learning framework for bert further pre-training

    Huan Li Xin’ao Wang, Ke Chen, and Lidan Shou. Fedbfpt: An efficient federated learning framework for bert further pre-training. In Proceedings of the Thirty-Second Inter- national Joint Conference on Artificial Intelligence , pages 4344–4352, 2023. 5, 6, 7

    2023

  25. [33]

    Weilong Yan, Ming Li, Haipeng Li, Shuwei Shao, and Robby T. Tan. Synthetic-to-real self-supervised robust depth estimation via learning with motion and structure priors. In Proceedings of the Computer Vision and Pattern Recognition Conference (CVPR), pages 21880–21890, 2025. 3

    2025

  26. [34]

    Bag-of-visual-words and spa- tial extensions for land-use classification

    Yi Yang and Shawn Newsam. Bag-of-visual-words and spa- tial extensions for land-use classification. In Proceedings of the 18th SIGSPATIAL international conference on advances in geographic information systems, pages 270–279, 2010. 5

    2010

  27. [35]

    See through gradients: Image batch recovery via gradinversion

    Hongxu Yin, Arun Mallya, Arash Vahdat, Jose M Alvarez, Jan Kautz, and Pavlo Molchanov. See through gradients: Image batch recovery via gradinversion. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 16337–16346, 2021. 4

    2021

  28. [36]

    Bitfit: Simple parameter-efficient fine-tuning for transformer-based masked language-models

    Elad Ben Zaken, Shauli Ravfogel, and Yoav Goldberg. Bitfit: Simple parameter-efficient fine-tuning for transformer-based masked language-models. arXiv preprint arXiv:2106.10199,

  29. [37]

    Fedpetuning: When fed- erated learning meets the parameter-efficient tuning methods of pre-trained language models

    Zhuo Zhang, Yuanhang Yang, Yong Dai, Qifan Wang, Yue Yu, Lizhen Qu, and Zenglin Xu. Fedpetuning: When fed- erated learning meets the parameter-efficient tuning methods of pre-trained language models. In Annual Meeting of the Association of Computational Linguistics 2023, pages 9963–

    2023

  30. [38]

    idlg: Improved deep leakage from gradients

    Bo Zhao, Konda Reddy Mopuri, and Hakan Bilen. idlg: Improved deep leakage from gradients. arXiv preprint arXiv:2001.02610, 2020. 4

    2001 arXiv

  31. [39]

    Favchat: Unlocking fine-grained facial video understanding with multimodal large language models

    Fufangchen Zhao, Ming Li, Linrui Xu, Wenhao Jiang, Jian Gao, and Danfeng Yan. Favchat: Unlocking fine-grained facial video understanding with multimodal large language models. arXiv preprint arXiv:2503.09158, 2025. 3

    2025 arXiv

  32. [40]

    Rethinking semantic segmen- tation from a sequence-to-sequence perspective with trans- formers

    Sixiao Zheng, Jiachen Lu, Hengshuang Zhao, Xiatian Zhu, Zekun Luo, Yabiao Wang, Yanwei Fu, Jianfeng Feng, Tao Xiang, Philip HS Torr, et al. Rethinking semantic segmen- tation from a sequence-to-sequence perspective with trans- formers. In Proceedings of the IEEE/CVF conference...

  33. [41]

    Deep leakage from gradients

    Ligeng Zhu, Zhijian Liu, and Song Han. Deep leakage from gradients. Advances in neural information processing sys- tems, 32, 2019. 4

    2019

  34. [42]

    Vistorybench: Comprehensive benchmark suite for story visualization

    Cailin Zhuang, Ailin Huang, Wei Cheng, Jingwei Wu, Yaoqi Hu, Jiaqi Liao, Zhewei Huang, Hongyuan Wang, Xinyao Liao, Weiwei Cai, et al. Vistorybench: Comprehensive benchmark suite for story visualization. arXiv preprint arXiv:2505.24862, 2025. 3

    2025

  35. [9977]

    Association for Computational Linguistics (ACL),

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.