Pith. sign in

REVIEW 4 major objections 4 minor 54 references

ViT-ProtoNet for Few-Shot Image Classification: A Multi-Benchmark Evaluation

T0 review · 4 major / 4 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read ViT-ProtoNet beats CNN few-shot baselines by up to 3.2%

desk verdict A straightforward ViT-Small + ProtoNet combination whose headline numbers look like pretraining memorization, not few-shot generalization; missing baselines and ablations leave the central claims unsupported. read the letter →

arxiv 2507.09299 v1 pith:PFPJP4OQ submitted 2025-07-12 cs.CV cs.AIcs.LG

Abdulvahap Mutlu , Şengül Doğan , Türker Tuncer This is my paper
classification cs.CVcs.AIcs.LG
keywords few-shotlearningvisiontransformerprototypicalnetworksmeta-learningViT-ProtoNetself-attentionimageclassificationbenchmarkevaluation
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

The paper proposes ViT-ProtoNet, which replaces the CNN encoder of a Prototypical Network with a ViT-Small vision transformer, and claims this simple swap improves 5-shot classification accuracy by up to 3.2 percentage points over CNN-based proto-nets. Across four benchmarks the reported accuracies are 96.60% on Mini-ImageNet, 96.53% on CUB-200, 95.25% on CIFAR-FS, and 81.88% on FC100, and the authors assert the model is the best ever trained on CIFAR-FS and FC100. The paper argues that self-attention features, averaged into class prototypes, give better class separability than local CNN features. If the claim holds, a lightweight pretrained transformer can serve as a practical, high-accuracy backbone for few-shot image classification.

What carries the argument

The machinery is the class prototype: the mean of the support samples' ViT feature embeddings, with queries classified by squared Euclidean distance to each prototype. The feature extractor is a ViT-Small/16 (patch size 16, 384-dim embeddings, 12 transformer layers, 6 heads), trained episodically for 1000 episodes with a prototypical negative-log-likelihood loss, AdamW with learning rate 1e-4, weight decay 1e-4, and dropout 0.1. Self-attention supplies global context that the prototype averaging then condenses into a single class center.

What would settle it

Find which dataset the ViT-Small weights were pretrained on (the paper does not say) and check whether any Mini-ImageNet, CIFAR-FS, or FC100 classes appear in it; if they do, retrain or re-extract features with a backbone pretrained only on disjoint classes and measure the accuracy gap — the central claim of few-shot generalization would be falsified if the gap over CNN baselines largely disappears.

Watch

Extended reading notes

Core claim

The central discovery is that averaging the CLS-token embeddings from a ViT-Small/16 backbone over the support set yields prototypes that classify novel classes with very high accuracy in the 5-way 5-shot setting. The paper reports 96.60% ± 0.54% on Mini-ImageNet, 96.53% ± 0.69% on CUB-200, 95.25% ± 0.82% on CIFAR-FS, and 81.88% ± 1.78% on FC100, surpassing CNN-based prototypical counterparts and, on CIFAR-FS and FC100, all prior models the authors compared against. The authors attribute the gain to the self-attention mechanism's ability to capture both global context and fine detail, which improves feature separability in latent space and makes the method robust to low-resolution and fine-grained inputs.

Load-bearing premise

The evaluation assumes the ViT-Small weights were pretrained on data that excludes the evaluation classes, yet Section 3.3 only says the authors 'extract features using a pre-trained ViT' without naming the pretraining corpus; because Mini-ImageNet and the CIFAR-100-derived benchmarks share classes with ImageNet, leakage could turn the reported few-shot gains into memorization.

Editorial extensions

If this is right

  • ViT-ProtoNet with a ViT-Small backbone sets a new reference point for transformer-based meta-learners on 5-way 5-shot classification, according to the paper.
  • The method beats all prior reported models on CIFAR-FS and FC100 in the 5-shot setting, suggesting self-attention features transfer well to low-resolution (32x32) inputs.
  • On fine-grained CUB-200, a ViT-Small model is competitive with ViT-Huge and Wide-ResNet baselines, indicating that large backbones are not necessary for this task.
  • CNN-based prototypical networks leave accuracy on the table; simply swapping the encoder for a pretrained ViT is a direct upgrade.
  • Training for only 1000 episodes on a single consumer GPU makes the approach computationally accessible and easy to reproduce.

Reading between the lines

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

  • The pretraining-corpus question is not addressed anywhere in the paper (Section 3.3 only mentions 'a pre-trained ViT'); if the backbone saw those classes, the reported accuracies would reflect memorization, not few-shot generalization.
  • The 1000-episode training budget is short compared with typical meta-training, implying most discriminative power comes from the pretrained features rather than the episodic objective, making the method closer to feature transfer than to meta-learning.
  • A natural experiment would freeze the ViT backbone and compare a plain nearest-prototype classifier against the episodically trained version to quantify what episodic training adds, if anything.
  • The pattern of gains, largest on CIFAR-100-derived datasets, is worth probing; it may indicate that the benchmark's small image size rather than the model's generality drives the improvement.
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 / 4 minor

Summary. This manuscript proposes ViT-ProtoNet, a few-shot classifier that replaces the convolutional backbone of Prototypical Networks with a ViT-Small/16 and uses the CLS token embedding as the image representation. Prototypes are computed as the mean support embedding per class, queries are classified by squared Euclidean distance, and training is episodic with the prototypical negative log-likelihood loss (Eqs. 1-4). The authors report 5-way 5-shot accuracies of 96.60% on Mini-ImageNet, 96.53% on CUB-200, 95.25% on CIFAR-FS, and 81.88% on FC100 (Table 1), compare with a small set of existing methods (Tables 2-5), and claim consistent superiority over CNN-based ProtoNets and state-of-the-art results on CIFAR-FS and FC100. The paper also promises code and pretrained weights, overlapped-support experiments, and ablations that do not appear in the body.

Significance. The proposed combination is straightforward and, if evaluated under clean protocols, would be a useful baseline: pretrained ViT features plus prototype averaging is simple, computationally light, and easy to re-implement. However, the paper's central empirical claims are not currently supported. The pretraining corpus is undisclosed despite Mini-ImageNet and CIFAR-FS/FC100 class overlap with ImageNet; no CNN ProtoNet baseline is run; and the external comparisons mix backbones, pretraining sources, and transductive/inductive settings. The reported near-saturated numbers are consistent with recognition of familiar classes rather than with few-shot generalization. Because the manuscript also fails to deliver the promised code link and several promised experiments, it cannot serve as a reliable benchmark contribution in its present form.

major comments (4)
  1. [Secs. 3.3, 3.5; Table 1] The ViT-Small backbone is described only as 'pre-trained' without identifying the pretraining corpus, the specific checkpoint, or whether the backbone is frozen or fine-tuned during episodic training. This is load-bearing because Mini-ImageNet is an ImageNet subset, CIFAR-FS and FC100 are CIFAR-100 subsets whose classes overlap ImageNet, and CUB-200 bird species substantially overlap ImageNet synsets. If an ImageNet-pretrained checkpoint was used, the evaluation classes have already been seen during pretraining, and the near-saturated accuracies in Table 1 could reflect recognition of familiar classes rather than few-shot generalization. The paper must disclose the exact checkpoint and pretraining data, verify that evaluation classes are excluded from pretraining, clarify the fine-tuning protocol, and include a control with a randomly initialized or otherwise non-overlapping backbone. The promised link to released code and weights is also absent, which compounds the audit problem.
  2. [Sec. 3.7, Tables 2-5] The abstract's headline claim that ViT-ProtoNet 'consistently outperforms its CNN-based prototypical counterparts' is not backed by any experiment in the manuscript: no CNN ProtoNet baseline appears in Tables 2-5. The listed comparators differ in backbone (Conv4, ResNet-12, WRN, ViT-Base/Huge), in pretraining (e.g., CAML uses a Laion-2b-based model), and in inference paradigm (Table 5 includes the transductive method BAVARDAGE). These comparisons cannot isolate the effect of the ViT backbone. A same-protocol comparison with Conv-4/ResNet-12 ProtoNets and with the same ViT-Small backbone, both with and without pretraining, is necessary to support the central claim.
  3. [Abstract; Secs. 3.4, 3.6, 4] The abstract promises evaluation of 'overlapped support variants' and 'comprehensive ablations' of transformer depth, patch size, and fine-tuning strategy, but the body does not report overlapped-support results, and the only ablation is a ViT-Tiny versus ViT-Small comparison in Section 3.4. There is no depth or patch-size sweep and no fine-tuning-strategy comparison, and Section 4.5 is missing entirely. These promised experiments are part of the stated contribution and should either be reported with full protocols or removed from the claims.
  4. [Sec. 2.6; Tables 2-5] The evaluation is too thin to support the claim in Section 4.1 of being 'the best model that has been ever trained on CIFAR-FS and FC100.' The protocol uses only 100 test episodes with 5-way 5-shot tasks per dataset, and the external results in Tables 2-5 are not matched in episode count, split version, or task sampling. The FC100 gain of 81.88% versus the next reported 70.60% is unusually large and would need to be confirmed with a much larger number of episodes and with the same evaluation script applied to the comparators before it can be taken as evidence of a methodological advance. In addition, Tables 2 and 3 show that SgVA-CLIP and CAML outperform ViT-ProtoNet on Mini-ImageNet and CUB-200, so the state-of-the-art wording should be restricted accordingly.
minor comments (4)
  1. [Sec. 2.5, Step 1] The augmentation description is malformed: 'randomly flipped horizontally and rotated (up to 10 of [0.5, 0.5, 0.5]' is incomplete and should specify rotation degrees and normalization constants explicitly.
  2. [Figures 1-2 and Sec. 3.4] Figure 2 appears twice with different captions ('Training Process' and 'Training Progress'), and Section 3.4 refers to a performance comparison as 'Figure 2' while the displayed figure appears to be a method diagram; renumber and retitle the figures.
  3. [Section 4] Section numbering skips from 4.4 to 4.6; renumber the subsections.
  4. [References] Some citations appear irrelevant or misplaced, for example reference 39 (Transflower) in the hybrid-approaches discussion; please verify all bibliography entries against the text.

Circularity Check

0 steps flagged · score 1.0 of 10

No circular derivation chain is exhibited; ViT-ProtoNet is an empirical benchmark evaluation whose prototype equations are standard ProtoNet definitions, so the central claims do not reduce to their inputs by construction.

full rationale

This is an empirical evaluation, not a derivation: the paper defines prototypes as support-embedding means (Eq. 1), squared-Euclidean classification (Eqs. 2-3), and the prototypical loss (Eq. 4), then measures accuracy on standard benchmarks. The prototype computation is definitionally the ProtoNet rule, not a fitted parameter later renamed as a prediction. No uniqueness theorem, ansatz, or load-bearing self-citation appears: the reference list contains no prior work by the authors, and the cited ViT paper is an architectural source. The principal validity risk is that the pretrained ViT-Small is never specified (Section 3.5: 'we extracted features using a pre-trained ViT'), and benchmarks such as Mini-ImageNet and the CIFAR-100 variants may overlap with common ViT pretraining corpora; that is a data-leakage and control concern, not a circular-equivalence concern, because the paper does not state the pretraining data and the reduction cannot be exhibited from the paper's text. Similarly, the ViT-Tiny vs ViT-Small choice in Section 3.4 may have been made using benchmark accuracy, which would be model-selection bias rather than a by-construction forcing of the reported numbers. Under the rule that circularity must be exhibited by quotation and explicit reduction, no qualifying circular step is found.

Assumptions & free parameters 8 free parameters · 3 assumptions · 0 invented entities

The method is an empirical pipeline resting on a pretrained transformer whose provenance is unspecified, on hand-chosen hyperparameters, and on the standard few-shot assumption that evaluation classes were unseen during pretraining. No new entities are introduced.

free parameters (8)
  • Pretrained ViT-Small initialization = unspecified checkpoint
    The method's performance depends on the pretrained weights, but the paper never identifies the checkpoint or its pretraining corpus. This is the largest hidden degree of freedom.
  • Number of training episodes = 1000
    Chosen by hand in Section 2.5 Step 7; no ablation or convergence analysis justifies this value.
  • Learning rate = 1e-4
    AdamW learning rate used in Section 3.3; no tuning reported.
  • Weight decay = 1e-4
    L2 regularization strength in Section 3.3; hand-chosen.
  • Dropout rate = 0.1
    Dropout applied in transformer layers per Section 3.2; no sensitivity analysis.
  • Batch size = 64 episodic tasks
    Batch size in Section 3.3; hand-chosen.
  • Image input resolution = 224x224
    All datasets, including 32x32 CIFAR-FS and FC100, are resized to 224x224; this choice affects feature quality and was not varied.
  • Evaluation episodes = 100
    Section 2.6 reports 100 test episodes repeated 5 times; this is lower than the common 600 or 10,000 episodes, affecting confidence interval stability.
assumptions (3)
  • domain assumption Evaluation classes are disjoint from the pretraining data of the ViT backbone.
    Section 3.3 uses a pretrained ViT without naming its pretraining data; Mini-ImageNet is a subset of ImageNet and CIFAR-100 classes overlap ImageNet, so this assumption is likely violated.
  • domain assumption Episodic training for 1000 episodes is sufficient to reach the reported performance.
    Section 2.5 fixes 1000 training episodes with no convergence check or ablation.
  • domain assumption The standard class-disjoint splits for the four benchmarks are used without modification.
    Section 2.6.1 describes datasets but not the exact splits; the abstract mentions 'overlapped support variants' that never appear in the text, so the evaluation protocol is incompletely specified.

how reviews work

0 comments
Cite this review

Pith. "Pith review of ViT-ProtoNet for Few-Shot Image Classification: A Multi-Benchmark Evaluation." pith.science (2026). https://pith.science/paper/PFPJP4OQ

@misc{pith2026250709299,
  author       = {Pith},
  title        = {Pith review of: ViT-ProtoNet for Few-Shot Image Classification: A Multi-Benchmark Evaluation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/PFPJP4OQ}},
  note         = {Machine review of arXiv:2507.09299}
}
read the original abstract

The remarkable representational power of Vision Transformers (ViTs) remains underutilized in few-shot image classification. In this work, we introduce ViT-ProtoNet, which integrates a ViT-Small backbone into the Prototypical Network framework. By averaging class conditional token embeddings from a handful of support examples, ViT-ProtoNet constructs robust prototypes that generalize to novel categories under 5-shot settings. We conduct an extensive empirical evaluation on four standard benchmarks: Mini-ImageNet, FC100, CUB-200, and CIFAR-FS, including overlapped support variants to assess robustness. Across all splits, ViT-ProtoNet consistently outperforms CNN-based prototypical counterparts, achieving up to a 3.2\% improvement in 5-shot accuracy and demonstrating superior feature separability in latent space. Furthermore, it outperforms or is competitive with transformer-based competitors using a more lightweight backbone. Comprehensive ablations examine the impact of transformer depth, patch size, and fine-tuning strategy. To foster reproducibility, we release code and pretrained weights. Our results establish ViT-ProtoNet as a powerful, flexible approach for few-shot classification and set a new baseline for transformer-based meta-learners.

Figures

Figures reproduced from arXiv: 2507.09299 by the authors.

Figure 1
Figure 1. Training Process a) The used ViT b) The present ViT-ProtoNet [PITH_FULL_IMAGE:figures/full_fig_p016_1.png] view at source ↗
Figure 2
Figure 2. Training Progress Our model, the loss optimization methods we used, and the ViT approach are shown in [PITH_FULL_IMAGE:figures/full_fig_p017_2.png] view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

54 extracted references · 28 canonical work pages

  1. [1]

    & Levine, S

    Finn, C., Abbeel, P. & Levine, S. Model-Agnostic Meta-Learning for Fast Adaptation of Deep Networks. Preprint at https://doi.org/10.48550/arXiv.1703.03400 (2017)

    arXiv doi:10.48550/arxiv.1703.03400 2017

  2. [2]

    & Lillicrap, T

    Vinyals, O., Blundell, C. & Lillicrap, T. Matching Networks for One Shot Learning

    work page

  3. [3]

    & Zemel, R

    Snell, J., Swersky, K. & Zemel, R. S. Prototypical Networks for Few-shot Learning. Preprint at https://doi.org/10.48550/arXiv.1703.05175 (2017)

    arXiv doi:10.48550/arxiv.1703.05175 2017

  4. [4]

    & Haffner, P

    Lecun, Y., Bottou, L., Bengio, Y. & Haffner, P. Gradient-based learning applied to document recognition. Proc. IEEE 86, 2278–2324 (1998)

    work page 1998

  5. [5]

    Dosovitskiy, A. et al. An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale. Preprint at https://doi.org/10.48550/arXiv.2010.11929 (2021)

    arXiv doi:10.48550/arxiv.2010.11929 2021

  6. [6]

    Vaswani, A. et al. Attention Is All You Need. Preprint at https://doi.org/10.48550/arXiv.1706.03762 (2023)

    arXiv doi:10.48550/arxiv.1706.03762 2023

  7. [7]

    CrossTransformers: spatially-aware few-shot transfer

    Doersch, C., Gupta, A. & Zisserman, A. CrossTransformers: spatially-aware few-shot transfer. Preprint at https://doi.org/10.48550/arXiv.2007.11498 (2021)

    work page Pith review arXiv doi:10.48550/arxiv.2007.11498 2021

  8. [8]

    F., Torr, P

    Bertinetto, L., Henriques, J. F., Torr, P. H. S. & Vedaldi, A. Meta-learning with differentiable closed-form solvers. Preprint at https://doi.org/10.48550/arXiv.1805.08136 (2019)

    arXiv doi:10.48550/arxiv.1805.08136 2019

Show all 54 references
  1. [9]

    The Caltech-UCSD Birds-200- 2011 Dataset

    Wah, C.; Branson, S.; Welinder, P.; Perona, P.; Belongie, S. The Caltech-UCSD Birds-200- 2011 Dataset. https://www.vision.caltech.edu/datasets/cub_200_2011/ (2011)

    2011

  2. [10]

    N., Rodriguez, P

    Oreshkin, B. N., Rodriguez, P. & Lacoste, A. TADAM: Task dependent adaptive metric for improved few-shot learning. Preprint at https://doi.org/10.48550/arXiv.1805.10123 (2019)

    2019 doi arXiv

  3. [11]

    Sung, F. et al. Learning to Compare: Relation Network for Few-Shot Learning. Preprint at https://doi.org/10.48550/arXiv.1711.06025 (2018)

    2018 doi arXiv

  4. [12]

    Chen, W.-Y., Liu, Y.-C., Kira, Z., Wang, Y.-C. F. & Huang, J.-B. A CLOSER LOOK AT FEW-SHOT CLASSIFICATION. (2019)

    2019

  5. [13]

    H., Shehata, M

    Walsh, R., Abdelpakey, M. H., Shehata, M. S. & Mohamed, M. M. Automated human cell classification in sparse datasets using few-shot learning. Sci. Rep. 12, 2924 (2022)

    2022

  6. [14]

    & Sun, J

    He, K., Zhang, X., Ren, S. & Sun, J. Deep Residual Learning for Image Recognition. Preprint at https://doi.org/10.48550/arXiv.1512.03385 (2015)

    2015 doi arXiv

  7. [15]

    van der & Weinberger, K

    Huang, G., Liu, Z., Maaten, L. van der & Weinberger, K. Q. Densely Connected Convolutional Networks. Preprint at https://doi.org/10.48550/arXiv.1608.06993 (2018)

    2018 doi arXiv

  8. [16]

    & Komodakis, N

    Zagoruyko, S. & Komodakis, N. Wide Residual Networks. Preprint at https://doi.org/10.48550/arXiv.1605.07146 (2017)

    2017 doi arXiv

  9. [17]

    & Drummond, T

    Hiller, M., Ma, R., Harandi, M. & Drummond, T. Rethinking Generalization in Few-Shot Classification

  10. [18]

    S., Chaudhari, P., Ravichandran, A

    Dhillon, G. S., Chaudhari, P., Ravichandran, A. & Soatto, S. A BASELINE FOR FEW- SHOT IMAGE CLASSIFICATION. (2020)

    2020

  11. [19]

    & Schulman, J

    Nichol, A., Achiam, J. & Schulman, J. On First-Order Meta-Learning Algorithms. Preprint at https://doi.org/10.48550/arXiv.1803.02999 (2018)

    2018 doi arXiv

  12. [20]

    Li, Z., Zhou, F., Chen, F. & Li, H. Meta-SGD: Learning to Learn Quickly for Few-Shot Learning. Preprint at https://doi.org/10.48550/arXiv.1707.09835 (2017)

    2017 doi arXiv

  13. [21]

    & Soatto, S

    Lee, K., Maji, S., Ravichandran, A. & Soatto, S. Meta-Learning with Differentiable Convex Optimization. Preprint at https://doi.org/10.48550/arXiv.1904.03758 (2019)

    2019 doi arXiv

  14. [22]

    Ren, M. et al. Meta-Learning for Semi-Supervised Few-Shot Classification. Preprint at https://doi.org/10.48550/arXiv.1803.00676 (2018)

    2018 doi arXiv

  15. [23]

    & Monfardini, G

    Scarselli, F., Gori, M., Ah Chung Tsoi, Hagenbuchner, M. & Monfardini, G. The Graph Neural Network Model. IEEE Trans. Neural Netw. 20, 61–80 (2009)

    2009

  16. [24]

    & Bruna, J

    Garcia, V. & Bruna, J. Few-Shot Learning with Graph Neural Networks. Preprint at https://doi.org/10.48550/arXiv.1711.04043 (2018)

    2018 doi arXiv

  17. [25]

    & Salakhutdinov, R

    Koch, G., Zemel, R. & Salakhutdinov, R. Siamese Neural Networks for One-shot Image Recognition

  18. [26]

    & Yoo, C

    Kim, J., Kim, T., Kim, S. & Yoo, C. D. Edge-labeling Graph Neural Network for Few- shot Learning. Preprint at https://doi.org/10.48550/arXiv.1905.01436 (2019)

    2019 doi arXiv

  19. [27]

    Liu, Y. et al. Learning to Propagate Labels: Transductive Propagation Network for Few- shot Learning. Preprint at https://doi.org/10.48550/arXiv.1805.10002 (2019)

    2019 doi arXiv

  20. [28]

    & Sha, F

    Ye, H.-J., Hu, H., Zhan, D.-C. & Sha, F. Few-Shot Learning via Embedding Adaptation with Set-to-Set Functions. Preprint at https://doi.org/10.48550/arXiv.1812.03664 (2021)

    2021 doi arXiv

  21. [29]

    Triantafillou, E. et al. Meta-Dataset: A Dataset of Datasets for Learning to Learn from Few Examples. Preprint at https://doi.org/10.48550/arXiv.1903.03096 (2020)

    2020 doi arXiv

  22. [30]

    & Luo, B

    Wang, X., Wang, X., Jiang, B. & Luo, B. Few-Shot Learning Meets Transformer: Unified Query-Support Transformers for Few-Shot Classification. Preprint at https://doi.org/10.48550/arXiv.2208.12398 (2022)

    2022 doi arXiv

  23. [31]

    Peng, F., Yang, X., Xiao, L., Wang, Y. & Xu, C. SgVA-CLIP: Semantic-guided Visual Adapting of Vision-Language Models for Few-shot Image Classification. Preprint at https://doi.org/10.48550/arXiv.2211.16191 (2023)

    2023 doi arXiv

  24. [32]

    Radford, A. et al. Learning Transferable Visual Models From Natural Language Supervision. Preprint at https://doi.org/10.48550/arXiv.2103.00020 (2021)

    2021 doi arXiv

  25. [33]

    & Zelnik-Manor, L

    Ridnik, T., Ben-Baruch, E., Noy, A. & Zelnik-Manor, L. ImageNet-21K Pretraining for the Masses. Preprint at https://doi.org/10.48550/arXiv.2104.10972 (2021)

    2021 doi arXiv

  26. [34]

    & Abbeel, P

    Mishra, N., Rohaninejad, M., Chen, X. & Abbeel, P. A Simple Neural Attentive Meta- Learner. Preprint at https://doi.org/10.48550/arXiv.1707.03141 (2018)

    2018 doi arXiv

  27. [35]

    Munkhdalai, T. & Yu, H. Meta Networks. Preprint at https://doi.org/10.48550/arXiv.1703.00837 (2017)

    2017 doi arXiv

  28. [36]

    M., Dolz, J., Granger, E

    Ziko, I. M., Dolz, J., Granger, E. & Ayed, I. B. Laplacian Regularized Few-Shot Learning. Preprint at https://doi.org/10.48550/arXiv.2006.15486 (2021)

    2021 doi arXiv

  29. [37]

    Goodfellow, I. J. et al. Generative Adversarial Networks. Preprint at https://doi.org/10.48550/arXiv.1406.2661 (2014)

    2014 doi arXiv

  30. [38]

    Kingma, D. P. & Welling, M. Auto-Encoding Variational Bayes. Preprint at https://doi.org/10.48550/arXiv.1312.6114 (2022)

    2022 doi arXiv

  31. [39]

    Valle-Pérez, G. et al. Transflower: probabilistic autoregressive dance generation with multimodal attention. ACM Trans. Graph. 40, 1–14 (2021)

    2021

  32. [40]

    & Wang, X

    Chen, Y., Liu, Z., Xu, H., Darrell, T. & Wang, X. Meta-Baseline: Exploring Simple Meta- Learning for Few-Shot Learning. Preprint at https://doi.org/10.48550/arXiv.2003.04390 (2021)

    2021 doi arXiv

  33. [41]

    & Salakhutdinov, R

    Srivastava, N., Hinton, G., Krizhevsky, A., Sutskever, I. & Salakhutdinov, R. Dropout: A Simple Way to Prevent Neural Networks from Overfitting

  34. [42]

    & Friedman, J

    Hastie, T., Tibshirani, R. & Friedman, J. H. The Elements of Statistical Learning: Data Mining, Inference, and Prediction. (Springer, New York, NY, 2017)

    2017

  35. [43]

    & Courville, A

    Goodfellow, I., Bengio, Y. & Courville, A. Deep Learning. (The MIT Press, Cambridge, Massachusetts, 2016)

    2016

  36. [44]

    & Hutter, F

    Loshchilov, I. & Hutter, F. Decoupled Weight Decay Regularization. Preprint at https://doi.org/10.48550/arXiv.1711.05101 (2019)

    2019 doi arXiv

  37. [45]

    He, T. et al. Bag of Tricks for Image Classification with Convolutional Neural Networks. Preprint at https://doi.org/10.48550/arXiv.1812.01187 (2018)

    2018 doi arXiv

  38. [46]

    Fifty, C. et al. Context-Aware Meta-Learning. Preprint at https://doi.org/10.48550/arXiv.2310.10971 (2024)

    2024 doi arXiv

  39. [47]

    X., Li, D., Stühmer, J., Kim, M

    Hu, S. X., Li, D., Stühmer, J., Kim, M. & Hospedales, T. M. Pushing the Limits of Simple Pipelines for Few-Shot Learning: External Data and Fine-Tuning Make a Difference. Preprint at https://doi.org/10.48550/arXiv.2204.07305 (2022)

    2022 doi arXiv

  40. [48]

    & Jamali-Rad, H

    Singh, A. & Jamali-Rad, H. Transductive Decoupled Variational Inference for Few-Shot Classification. Preprint at https://doi.org/10.48550/arXiv.2208.10559 (2022)

    2022 doi arXiv

  41. [49]

    & Korman, S

    Shalam, D. & Korman, S. The Self-Optimal-Transport Feature Transform

  42. [50]

    & Korman, S

    Shalam, D. & Korman, S. The Balanced-Pairwise-Affinities Feature Transform. Preprint at https://doi.org/10.48550/arXiv.2407.01467 (2024)

    2024 doi arXiv

  43. [51]

    & Pateux, S

    Hu, Y., Gripon, V. & Pateux, S. Squeezing Backbone Feature Distributions to the Max for Efficient Few-Shot Learning. Preprint at https://doi.org/10.48550/arXiv.2110.09446 (2021)

    2021 doi arXiv

  44. [52]

    & Gripon, V

    Hu, Y., Pateux, S. & Gripon, V. Adaptive Dimension Reduction and Variational Inference for Transductive Few-Shot Classification. Preprint at https://doi.org/10.48550/arXiv.2209.08527 (2022)

    2022 doi arXiv

  45. [53]

    & Lin, C.-M

    Liu, J., Chao, F. & Lin, C.-M. Task Augmentation by Rotating for Meta-Learning. Preprint at https://doi.org/10.48550/arXiv.2003.00804 (2020)

    2020 doi arXiv

  46. [54]

    & Qin, Y

    Song, L., Liu, J. & Qin, Y. Generalized Adaptation for Few-Shot Learning. Preprint at https://doi.org/10.48550/arXiv.1911.10807 (2020). Appendix Training Codes Attention Map t-SNE Visualization

    2020 doi arXiv

Pith tools

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