Pith. sign in

REVIEW 5 major objections 6 minor 77 references

This paper claims that swapping the KL divergence for the Hellinger distance in variational few-shot inference yields a network that is more accurate and more resistant to adversarial and natural noise, backed by a Hellinger-based ELBO.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

ANROT-HELANet combines Hellinger aggregation, attention, and FGSM/Gaussian robust training for few-shot classification, but its ELBO derivation is invalid and its performance claims are overstated.

T0 review reviewed 2026-08-04 challenge →

load-bearing objection The empirical recipe is a real extension of the authors' HELA-VFA line, but the core theory is invalid: the Hellinger ELBO derivation fails at two algebraic steps, and the training loss never uses the derived objective. the 5 major comments →

arxiv 2509.11220 v1 pith:MILG6HHH submitted 2025-09-14 cs.CV

ANROT-HELANet: Adverserially and Naturally Robust Attention-Based Aggregation Network via The Hellinger Distance for Few-Shot Classification

Gao Yu Lee , Tanmoy Dam , Md Meftahul Ferdaus , Daniel Puiu Poenar , Vu N.Duong This is my paper
classification cs.CV
keywords few-shot learningHellinger distancevariational inferenceevidence lower boundadversarial robustnessnatural robustnessattention mechanismcontrastive loss
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

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 sets out to establish that the Hellinger distance can replace the Kullback-Leibler divergence as the divergence used in variational few-shot classification, providing a bounded, symmetric divergence that behaves better when data are scarce or perturbed. The proposed network, ANROT-HELANet, uses attention-based feature extraction, Hellinger-distance class-prototype aggregation, and a Hellinger similarity softmax loss, and is trained jointly on FGSM adversarial examples and Gaussian-corrupted images. The authors derive an evidence-lower-bound-like objective, ELBO' = ∫ log pθ(T) dz + log(1 − D_H^2)^2, and argue this follows from rewriting the ELBO with Hellinger distance. If that derivation holds, few-shot models could resist adversarial perturbations up to epsilon=0.30 and Gaussian noise up to sigma=0.30 while improving accuracy on the four benchmarks tested, and the Hellinger similarity loss would generalize contrastive cosine-similarity losses. The paper also reports better reconstructed-image quality for the Hellinger-based variational autoencoder, with FID 2.75 versus 3.43 for the KL-based and 3.38 for the Wasserstein-based versions.

Core claim

The paper's central claim is that the Hellinger distance, D_H^2 = 1 − ∫ sqrt(pθ(z|T) qφ(z|S)) dz, can form the basis of a valid evidence lower bound for variational few-shot inference. The authors derive ELBO' = ∫ log pθ(T) dz + log(1 − D_H^2)^2, with ELBO' written as ELBO divided by qφ(z|S), and argue that D_H's symmetry, [0,1] bound, closed-form Gaussian expression, and relation to the Bhattacharyya/Mahalanobis distances yield smoother gradients and steadier meta-training than the asymmetric, unbounded KL divergence. Around this they build ANROT-HELANet: attention-based ResNet-12 encoder, Hellinger prototype aggregation, a Hellinger similarity softmax loss, and training on FGSM adversarial

What carries the argument

The central object is the Hellinger distance D_H(P,Q) = sqrt(1 − ∫√(p q) dx), a symmetric f-divergence bounded in [0,1] that acts like a probabilistic Euclidean distance. It carries the argument by replacing the KL divergence in the variational objective: the paper's Eq (21) rewrites the ELBO as ELBO' = ∫ log pθ(T) dz + log(1 − D_H^2)^2, and the same distance defines the Hellinger similarity loss L_Hesim used in the softmax over class prototypes. The closed-form Gaussian form of D_H, along with its link to the Bhattacharyya coefficient and Mahalanobis distance, lets the model compute class-prototype distances directly and, the paper argues, keeps gradient norms moderate when distributions on

Load-bearing premise

The derivation's load-bearing premise is that the logarithm can be moved inside the integral in Eqs (5)-(7) and that the ELBO can be divided by the density qφ(z|S) in Eq (20); these steps are not generally valid, so if they fail, the theoretical foundation for the Hellinger-based ELBO collapses and only the empirical pipeline remains.

What would settle it

Compute Eq (7) for two unit-variance Gaussian densities with different means: the left side log∫√(p q) dz is a finite number, while the right side ∫log(p q) dz diverges to −∞ because the log of a Gaussian density integrates to a divergent negative quantity. This one calculation would show the derived Hellinger ELBO is not a consequence of the stated equations. A complementary check is to compute the exact ELBO for a synthetic two-Gaussian model and compare it against Eq (21).

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • If the Hellinger ELBO is valid, variational few-shot inference can be optimized with a bounded, symmetric divergence, which should remove the asymmetry artifacts of KL and make updates more consistent across episodes.
  • Training with FGSM and Gaussian-corrupted samples together gives a single model that retains 66.0% accuracy at adversarial epsilon=0.30 and 84.1% at Gaussian sigma=0.30 on miniImageNet 5-shot, whereas the same model without robustness training drops to 50.5% at epsilon=0.10.
  • The Hellinger similarity loss can replace cosine similarity in contrastive few-shot losses and improves prototype aggregation, as shown by the ablation study where attention plus L_Hesim outperforms attention plus KL.
  • Hellinger-based VAE reconstruction gives a lower FID (2.75) than KL-based (3.43) and Wasserstein-based (3.38) VAEs, indicating better generative quality for the same backbone.
  • Attention maps stay on discriminative object regions under small perturbations, so the attention mechanism contributes to robustness and can be visualized to explain which features the model relies on.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Editorial inference: the reported gains are not a clean test of the Hellinger ELBO because the ablation changes several components at once (attention, loss function, noise training). A sharper experiment would fix one architecture and compare KL-based and Hellinger-based objectives under identical augmentation and attention settings.
  • Editorial inference: if the derivation in Eqs (5)-(21) is set aside, the boundedness of Hellinger distance still suggests a usable property: because the distance cannot exceed 1, its gradient cannot blow up as easily as KL's, so the empirical robustness may transfer to other symmetric f-divergences such as Bhattacharyya distance.
  • Editorial inference: the closed-form Gaussian Hellinger distance is cheap to compute, which makes the aggregation scheme a natural fit for continual or online few-shot learning where class prototypes must be updated incrementally from few examples.
  • Editorial inference: the paper's reported feature-space displacements (2.3 units for adversarial vs 0.8 for Gaussian noise) imply a testable relationship: if Hellinger distance is truly the operative similarity, the accuracy gap under the two noise types should be predictable from the Hellinger distance between clean and perturbed class-conditional Gaussians.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

5 major / 6 minor

Summary. The paper proposes ANROT-HELANet, a few-shot classification method that combines an attention mechanism, FGSM-based adversarial training, Gaussian-noise augmentation, and Hellinger-distance-based feature aggregation and loss. The central theoretical claim, developed in Section 3.3, is that the Hellinger distance can replace the KL divergence in variational few-shot inference, yielding an ELBO-type objective ELBO' = ∫ log pθ(T) dz + log(1−D_H²)² (Eq. 21). The method is evaluated on CIFAR-FS, FC-100, miniImageNet, and tieredImageNet for 1-shot and 5-shot accuracy under adversarial and natural perturbations, plus reconstruction quality measured by FID.

Significance. If the theoretical and empirical claims were valid, the paper would be a useful contribution: it would extend variational few-shot learning to a symmetric, bounded divergence and combine it with attention-based adversarial/natural robustness. The empirical study is broad: four benchmark datasets, ablations over the proposed components, robustness curves over ǫ and σ, and a computational complexity comparison. The authors also state an intention to release code. However, the central mathematical derivation is invalid, the derived objective is not the one used in training, the abstract's state-of-the-art claim is contradicted by the paper's own tables, and the FID comparison is confounded by architecture. The remaining empirical recipe may be of interest, but the paper's central claims as written are not supported.

major comments (5)
  1. [Section 3.3, Eqs. (5)-(7)] The derivation moves the logarithm inside an integral, asserting log ∫ sqrt(pθ qφ) dz = ∫ log(pθ qφ) dz. This is false in general; Jensen's inequality gives log E[·] ≥ E[log ·]. For a concrete counterexample, take pθ=qφ=N(0,1): the left side of Eq. (5) is log 1 = 0, while the right side of Eq. (7) is 2∫ log φ(z) dz = −log(2π)−1 ≠ 0. Since Eq. (21) is built on this step, the claimed Hellinger-ELBO relation does not follow.
  2. [Section 3.3, Eq. (20)] The asserted identity ∫ log(pθ(z,T)/qφ(z|S)) dz = (1/qφ(z|S))·ELBO is not mathematically valid: the ELBO is a scalar, qφ(z|S) is a density, and the intermediate expression ∫ (1/qφ(z|S))·(dELBO/dz) dz is undefined because differentiating a scalar ELBO with respect to the integration variable z is not well posed. Consequently ELBO' = ELBO/qφ(z|S) is not a scalar lower bound, and the comparison with Eq. (22) is also dimensionally inconsistent because ∫ log pθ(T) dz is an integral over z, not the log-evidence term.
  3. [Sections 3.3-3.4 and Algorithm 1] The derived objective in Eq. (21) is not the objective optimized by the network. The actual training loss is Eq. (30), L_HELANet = L_CCE + λ1·L_Hesim + λ2·L_rec. Neither L_CCE nor L_rec contains ELBO' or the Hellinger distance; L_Hesim (Eqs. 27-28) is a softmax over a similarity function. Thus even a corrected Hellinger-ELBO bound would not justify the training loss used, and the theoretical section is decoupled from the method.
  4. [Abstract and Section 5.1, Tables 4-5] The abstract claims ANROT-HELANet 'establishes new state-of-the-art performance,' but the paper's own experiments contradict this. On miniImageNet (Table 4), TRIDENT achieves 86.1±0.6 (1-shot) and 96.0±0.3 (5-shot) versus the proposed 69.4±0.3 and 88.1±0.4. On tieredImageNet (Table 5), TRIDENT achieves 87.0±0.6 and 97.0±0.2 versus 75.3±0.2 and 89.5±0.8. The text in Section 5.1 acknowledges that TRIDENT reports more superior metrics on these datasets. The state-of-the-art claim is load-bearing and contradicted by the reported data.
  5. [Section 5.2, Table 7] The FID comparison (2.75 for ANROT-HELANet vs. 3.43 and 3.38 for the baselines) compares different architectures: ANROT-HELANet includes attention, adversarial/Gaussian training, and multiple auxiliary losses, whereas the baselines are vanilla VAEs/WAE. The improvement cannot be attributed to the Hellinger distance without matched ablations that vary only the divergence. As reported, the reconstruction-quality claim is confounded and unsupported.
minor comments (6)
  1. [Title and Abstract] 'Adverserially' should be 'Adversarially' in the title and abstract.
  2. [Front matter] The PACS: 0000, 1111 and 2000 MSC: 0000, 1111 lines appear to be placeholder values and should be removed or filled.
  3. [Section 5.1.2 and Table 6] In the σ discussion, the prose refers to 'ǫ' instead of 'σ' in several places. Also, the 'Without Adversarial or Gaussian Training' rows of Table 6 list σ values 0.18, 0.06, 0.23, 0.33, 0.53, 0.41 instead of the stated 0.05, 0.10, 0.15, 0.20, 0.25, 0.30, and the 'With' rows list σ = 0.30 twice while omitting 0.25.
  4. [Equation (28)] The notation c(Q,j) is defined as the j-th class prototype from the query set; in standard few-shot classification, prototypes should be computed from the support set. This is either a typo or a methodological inconsistency.
  5. [Equation (33)] The WAE ELBO is written as qφ(z|S)log pθ(z|T) − λ D_W(...); this is missing the expectation over qφ and does not match the standard Wasserstein autoencoder objective in the cited references.
  6. [Figure 4 caption] The caption says 'Vanilla VAE architectures using the Hellinger distance,' but the model shown is ANROT-HELANet. The caption should be corrected to avoid implying that a vanilla VAE was used.

Circularity Check

1 steps flagged

No circular reduction in the empirically evaluated pipeline; the Hellinger-ELBO 'derivation' is definitional/invalid and disconnected from the training loss, and the headline gains are relative to the authors' own HELA-VFA baseline.

specific steps
  1. self definitional [Section 3.3 (ELBO and Hellinger Distance), Eqs. (20)-(21)]
    "Finally, noting the observation that ∫ log(pθ(z,T)/qφ(z|S)) dz can be rewritten in terms of ELBO and qφ(z|S) as ∫ (1/qφ(z|S) d(ELBO)/dz) dz = 1/qφ(z|S) ELBO, the following is derived: ELBO′ = ∫ log(pθ(T)) dz + log(1−D²_H)² where ELBO′ = ELBO/qφ(z|S)"

    The result called a 'derived' ELBO-type objective is created by definition: ELBO′ is stipulated to be ELBO/qφ(z|S), and the 'observation' merely rewrites the ELBO integrand divided by the density qφ. No independent variational bound is obtained; the Hellinger term is attached by algebraic manipulation that also moves log inside the integral (Eq. 5→7), which is not valid. Moreover, the actual training loss in Eq. (30), LHELANet = LCCE + λ1·LHesim + λ2·Lrec, contains no ELBO′ term, so this definitional 'result' is not what the network optimizes.

full rationale

The empirical claims are self-contained against external benchmarks: accuracies are measured on held-out meta-test tasks and compared with many external methods (ProtoNet, FeLMi, TRIDENT, etc.), and the ablations (Tables 8-11) directly replace Hellinger-based loss with KL or no similarity loss. No reported test accuracy is a fitted parameter renamed as a prediction. The main 'gains' quoted in the abstract (1.20%, 1.40% on miniImageNet) are, however, relative to the authors' own HELA-VFA [56], and the external TRIDENT baseline actually reports higher accuracy on miniImageNet (86.1/96.0 vs 69.4/88.1) and tieredImageNet (87.0/97.0 vs 75.3/89.5). This is a claim-calibration and self-citation issue, not a circular derivation. The theoretical Hellinger-ELBO section is mathematically unsound (log moved inside an integral; a scalar ELBO divided by a density), so the claimed first-principles result is unsupported, but it is also not load-bearing for the trained objective because Eq. (30) does not optimize ELBO′. Overall, the central evaluated system is not circular; the identified circular/definitional element is confined to the theory section that is disconnected from the actual training loss.

Axiom & Free-Parameter Ledger

3 free parameters · 5 axioms · 0 invented entities

The empirical part depends on a handful of hand-set hyperparameters and on an unverified assumption about feature-space FGSM. The theoretical part rests on two invalid algebraic steps, which are not axioms pulled from the literature but incorrect manipulations unique to this paper.

free parameters (3)
  • λ1 (Hellinger similarity loss weight) = 0.5
    Set in §3.4; governs balance of LHesim relative to cross-entropy, affects final accuracy in ablations.
  • λ2 (reconstruction loss weight) = 1.0
    Set in §3.4; affects reconstruction quality and training dynamics.
  • Training perturbation levels ǫ and σ = 0.05 and 0.05
    All reported models were trained with ǫ=0.05 and σ=0.05 (§5.1); robustness at larger test-time values depends on these chosen training noise levels.
axioms (5)
  • ad hoc to paper log of an integral equals integral of log
    Used to transform Eq (5) into Eq (7); false as a general identity.
  • ad hoc to paper Eq (20) identity: ∫ log(p/q) dz = (1/q) ELBO
    Defines ELBO' but is mathematically undefined; a scalar divided by a density.
  • standard math Closed-form Hellinger distance between Gaussians (from Pardo [41])
    Standard result used in Eq (23)-(24).
  • domain assumption FGSM perturbation of attention feature maps produces valid adversarial images
    Eq (3) and Algorithm 1; the paper never explains how perturbed feature maps are mapped back to images.
  • domain assumption Training at σ=0.05 confers robustness at test σ up to 0.30
    Asserted in §5.1.2 without a study varying training noise.

reviewed 2026-08-04 · how reviews work

0 comments
Cite this review

Pith. "Pith review of ANROT-HELANet: Adverserially and Naturally Robust Attention-Based Aggregation Network via The Hellinger Distance for Few-Shot Classification." pith.science (2026). https://pith.science/paper/MILG6HHH

@misc{pith2026250911220,
  author       = {Pith},
  title        = {Pith review of: ANROT-HELANet: Adverserially and Naturally Robust Attention-Based Aggregation Network via The Hellinger Distance for Few-Shot Classification},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/MILG6HHH}},
  note         = {Machine review of arXiv:2509.11220}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Few-Shot Learning (FSL), which involves learning to generalize using only a few data samples, has demonstrated promising and superior performances to ordinary CNN methods. While Bayesian based estimation approaches using Kullback-Leibler (KL) divergence have shown improvements, they remain vulnerable to adversarial attacks and natural noises. We introduce ANROT-HELANet, an Adversarially and Naturally RObusT Hellinger Aggregation Network that significantly advances the state-of-the-art in FSL robustness and performance. Our approach implements an adversarially and naturally robust Hellinger distance-based feature class aggregation scheme, demonstrating resilience to adversarial perturbations up to $\epsilon=0.30$ and Gaussian noise up to $\sigma=0.30$. The network achieves substantial improvements across benchmark datasets, including gains of 1.20\% and 1.40\% for 1-shot and 5-shot scenarios on miniImageNet respectively. We introduce a novel Hellinger Similarity contrastive loss function that generalizes cosine similarity contrastive loss for variational few-shot inference scenarios. Our approach also achieves superior image reconstruction quality with a FID score of 2.75, outperforming traditional VAE (3.43) and WAE (3.38) approaches. Extensive experiments conducted on four few-shot benchmarked datasets verify that ANROT-HELANet's combination of Hellinger distance-based feature aggregation, attention mechanisms, and our novel loss function establishes new state-of-the-art performance while maintaining robustness against both adversarial and natural perturbations. Our code repository will be available at https://github.com/GreedYLearner1146/ANROT-HELANet/tree/main.

Figures

Figures reproduced from arXiv: 2509.11220 by Daniel Puiu Poenar, Gao Yu Lee, Md Meftahul Ferdaus, Tanmoy Dam, Vu N.Duong.

Figure 1
Figure 1. Figure 1: Some illustrations of how even a small adversarial [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Algorithmic architecture of our attention mechan [PITH_FULL_IMAGE:figures/full_fig_p012_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Algorithmic architecture of our overall ANROT-HE [PITH_FULL_IMAGE:figures/full_fig_p025_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: The miniImageNet reconstructed images obtained fr [PITH_FULL_IMAGE:figures/full_fig_p026_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Comparative plots of ANROT-HELANet classificatio [PITH_FULL_IMAGE:figures/full_fig_p033_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: The miniImageNet reconstructed images obtained fr [PITH_FULL_IMAGE:figures/full_fig_p034_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: The miniImageNet reconstructed images obtained fr [PITH_FULL_IMAGE:figures/full_fig_p034_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: The miniImageNet reconstructed images obtained fr [PITH_FULL_IMAGE:figures/full_fig_p035_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: The original miniImageNet images. (zoom in for the best view). 35 [PITH_FULL_IMAGE:figures/full_fig_p035_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Plot of GRAD-CAM on the samples of the miniImageNe [PITH_FULL_IMAGE:figures/full_fig_p037_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Plot of GRAD-CAM on the samples of the miniImageNe [PITH_FULL_IMAGE:figures/full_fig_p038_11.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

77 extracted references · 3 canonical work pages · 3 internal anchors

  1. [1]

    G. Y . Lee, T. Dam, M. M. Ferdaus, D. P . Poenar, V . N. Duong, U nlocking the capabilities of explainable few-shot learning in remote sensing, Artificial Intelligence Review 57 (7) (2024) 169

    2024

  2. [2]

    G. Koch, R. Zemel, R. Salakhutdinov, et al., Siamese neur al networks for one-shot image recognition, in: ICML deep learning worksho p, V ol. 2, Lille, 2015

    2015

  3. [3]

    Snell, K

    J. Snell, K. Swersky, R. Zemel, Prototypical networks fo r few-shot learning, Advances in neural information processing systems 30 (2017 )

    2017

  4. [4]

    Zhang, C

    J. Zhang, C. Zhao, B. Ni, M. Xu, X. Y ang, V ariational few-s hot learning, in: Proceedings of the IEEE/CVF International Conference on Co mputer Vi- sion, 2019, pp. 1685–1694

    2019

  5. [5]

    E. Hellinger, Neue begr¨ undung der theorie quadratisch er formen von un- endlichvielen ver¨ anderlichen., Journal f¨ ur die reine und angewandte Mathe- matik 1909 (136) (1909) 210–271

    1909

  6. [6]

    Grzyb, J

    J. Grzyb, J. Klikowski, M. Wo´ zniak, Hellinger distance weighted ensemble for imbalanced data stream classification, Journal of Compu tational Science 51 (2021) 101314

    2021

  7. [7]

    Kumari, U

    A. Kumari, U. Thakar, Hellinger distance based oversamp ling method to solve multi-class imbalance problem, in: 2017 7th Internat ional Confer- ence on Communication Systems and Network Technologies (CSNT), IEEE, 2017, pp. 137–141

    2017

  8. [8]

    Kyrkou, T

    C. Kyrkou, T. Theocharides, Emergencynet: Efficient aer ial image classi- fication for drone-based emergency monitoring using atrous convolutional feature fusion, IEEE Journal of Selected Topics in Applied E arth Observa- tions and Remote Sensing 13 (2020) 1687–1699

    2020

  9. [9]

    Goldblum, L

    M. Goldblum, L. Fowl, T. Goldstein, Adversarially robus t few-shot learn- ing: A meta-learning approach, Advances in Neural Informat ion Processing Systems 33 (2020) 17886–17895

    2020

  10. [10]

    J. Dong, Y . Wang, J.-H. Lai, X. Xie, Improving adversari ally robust few- shot image classification with generalizable representati ons, in: Proceedings 46 of the IEEE/CVF Conference on Computer Vision and Pattern Re cognition, 2022, pp. 9025–9034

    2022

  11. [11]

    S. Wu, J. Sang, K. Xu, J. Zhang, J. Y u, Attention, please! adversar- ial defense via activation rectification and preservation, arXiv preprint arXiv:1811.09831 (2018)

    Pith/arXiv arXiv 2018

  12. [12]

    H. Wang, Y . Deng, S. Y oo, H. Ling, Y . Lin, Agkd-bml: Defense against ad- versarial attack by attention guided knowledge distillati on and bi-directional metric learning, in: Proceedings of the IEEE/CVF Internati onal Conference on Computer Vision, 2021, pp. 7658–7667

    2021

  13. [13]

    Agrawal, N

    P . Agrawal, N. S. Punn, S. K. Sonbhadra, S. Agarwal, Impa ct of attention on adversarial robustness of image classification models, in: 2021 IEEE Inter- national Conference on Big Data (Big Data), IEEE, 2021, pp. 3 013–3019

    2021

  14. [14]

    Hsieh, M

    Y .-L. Hsieh, M. Cheng, D.-C. Juan, W. Wei, W.-L. Hsu, C.- J. Hsieh, On the robustness of self-attentive models, in: Proceedings o f the 57th Annual Meeting of the Association for Computational Linguistics, 2019, pp. 1520– 1529

    2019

  15. [15]

    N. Mu, D. Wagner, Defending against adversarial patche s with robust self- attention, in: ICML 2021 Workshop on Uncertainty and Robust ness in Deep Learning, V ol. 1, 2021

    2021

  16. [16]

    Z. Li, C. Feng, M. Wu, H. Y u, J. Zheng, F. Zhu, Adversarial robustness via attention transfer, Pattern Recognition Letters 146 (2021 ) 172–178

    2021

  17. [17]

    Goodman, X

    D. Goodman, X. Li, J. Liu, D. Dou, T. Wei, Improving adver sarial robustness via attention and adversarial logit pairing, arXiv preprin t arXiv:1908.11435 (2019)

    Pith/arXiv arXiv 1908

  18. [18]

    C. Liu, Y . Dong, W. Xiang, X. Y ang, H. Su, J. Zhu, Y . Chen, Y. He, H. Xue, S. Zheng, A comprehensive study on robustness of image classification mod- els: Benchmarking and rethinking, arXiv preprint arXiv:23 02.14301 (2023)

    2023

  19. [19]

    Barbu, D

    A. Barbu, D. Mayo, J. Alverio, W. Luo, C. Wang, D. Gutfreu nd, J. Tenen- baum, B. Katz, Objectnet: A large-scale bias-controlled da taset for pushing the limits of object recognition models, Advances in neural information pro- cessing systems 32 (2019). 47

    2019

  20. [20]

    Hendrycks, S

    D. Hendrycks, S. Basart, N. Mu, S. Kadavath, F. Wang, E. D orundo, R. De- sai, T. Zhu, S. Parajuli, M. Guo, et al., The many faces of robu stness: A critical analysis of out-of-distribution generalization , in: Proceedings of the IEEE/CVF International Conference on Computer Vision, 202 1, pp. 8340– 8349

  21. [21]

    Hendrycks, T

    D. Hendrycks, T. Dietterich, Benchmarking neural netw ork robustness to common corruptions and perturbations, arXiv preprint arXi v:1903.12261 (2019)

    Pith/arXiv arXiv 1903

  22. [22]

    Mazumder, P

    P . Mazumder, P . Singh, V . P . Namboodiri, Rnnp: A robust few-shot learning approach, in: Proceedings of the IEEE/CVF Winter Conferenc e on Applica- tions of Computer Vision, 2021, pp. 2664–2673

    2021

  23. [23]

    T. Chen, S. Kornblith, M. Norouzi, G. Hinton, A simple fr amework for con- trastive learning of visual representations, in: Internat ional conference on machine learning, PMLR, 2020, pp. 1597–1607

    2020

  24. [24]

    Krizhevsky, G

    A. Krizhevsky, G. Hinton, et al., Learning multiple lay ers of features from tiny images (2009)

    2009

  25. [25]

    Bertinetto, J

    L. Bertinetto, J. F. Henriques, P . H. Torr, A. V edaldi, M eta-learning with differentiable closed-form solvers, arXiv preprint arXiv :1805.08136 (2018)

    Pith/arXiv arXiv 2018

  26. [26]

    Vinyals, C

    O. Vinyals, C. Blundell, T. Lillicrap, D. Wierstra, et a l., Matching networks for one shot learning, Advances in neural information proce ssing systems 29 (2016)

    2016

  27. [27]

    M. Ren, E. Triantafillou, S. Ravi, J. Snell, K. Swersky, J . B. Tenenbaum, H. Larochelle, R. S. Zemel, Meta-learning for semi-supervi sed few-shot classification, arXiv preprint arXiv:1803.00676 (2018)

    Pith/arXiv arXiv 2018

  28. [28]

    J. Han, Y . Ren, J. Ding, K. Y an, G.-S. Xia, Few-shot objec t detection via variational feature aggregation, arXiv preprint arXiv:23 01.13411 (2023)

    2023

  29. [29]

    Singh, H

    A. Singh, H. Jamali-Rad, Transductive decoupled varia tional inference for few-shot classification, arXiv preprint arXiv:2208.10559 (2022)

    Pith/arXiv arXiv 2022

  30. [30]

    Nguyen, C

    Q.-H. Nguyen, C. Q. Nguyen, D. D. Le, H. H. Pham, M. N. Do, E nhanc- ing few-shot image classification with cosine transformer, arXiv preprint arXiv:2211.06828 (2022). 48

    Pith/arXiv arXiv 2022

  31. [31]

    M. Chen, J. Du, R. Pasunuru, T. Mihaylov, S. Iyer, V . Stoyanov, Z. Kozareva, Improving in-context few-shot learning via self-supervis ed training, ArXiv abs/2205.01703 (2022). doi:10.48550/arXiv.2205.01703

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2205.01703 2022

  32. [32]

    Z. Y an, Y . An, H. Xue, Reinforced self-supervised train ing for few- shot learning, IEEE Signal Processing Letters 31 (2024) 731 –735. doi:10.1109/LSP .2024.3370488

    arXiv 2024

  33. [33]

    I. J. Goodfellow, J. Shlens, C. Szegedy, Explaining and harnessing adversar- ial examples, arXiv preprint arXiv:1412.6572 (2014)

    Pith/arXiv arXiv 2014

  34. [34]

    Y . Fu, Y . Xie, Y . Fu, Y . Jiang, Styleadv: Meta style adver sarial train- ing for cross-domain few-shot learning, ArXiv abs/2302.09 309 (2023). doi:10.48550/arXiv.2302.09309

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2302.09309 2023

  35. [35]

    J. Dong, Y . Wang, X. Xie, J. Lai, Y . Ong, Generalizable an d discrim- inative representations for adversarially robust few-sho t learning, IEEE Transactions on Neural Networks and Learning Systems PP (20 24). doi:10.1109/TNNLS.2024.3379172

    arXiv 2024

  36. [36]

    G. K. Nayak, R. Rawal, I. Khatri, A. Chakraborty, Robust few-shot learn- ing without using any adversarial samples, ArXiv abs/2211. 01598 (2022). doi:10.48550/arXiv.2211.01598

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2211.01598 2022

  37. [37]

    Madry, A

    A. Madry, A. Makelov, L. Schmidt, D. Tsipras, A. Vladu, T owards deep learning models resistant to adversarial attacks, arX iv preprint arXiv:1706.06083 (2017)

    Pith/arXiv arXiv 2017

  38. [38]

    C. Xie, Y . Wu, L. v. d. Maaten, A. L. Y uille, K. He, Featuredenoising for im- proving adversarial robustness, in: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2019, pp. 501–5 09

    2019

  39. [39]

    X. Y u, B. Han, J. Y ao, G. Niu, I. Tsang, M. Sugiyama, How do es disagree- ment help generalization against label corruption?, in: In ternational Confer- ence on Machine Learning, PMLR, 2019, pp. 7164–7173

    2019

  40. [40]

    H. Wei, L. Feng, X. Chen, B. An, Combating noisy labels by agree- ment: A joint training method with co-regularization, in: P roceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2020, pp. 13726–13735. 49

    2020

  41. [41]

    Pardo, Statistical inference based on divergence me asures, CRC press, 2018

    L. Pardo, Statistical inference based on divergence me asures, CRC press, 2018

    2018

  42. [42]

    G. J. McLachlan, Mahalanobis distance, Resonance 4 (6) (1999) 20–26

    1999

  43. [43]

    Bhattacharyya, On a measure of divergence between tw o multinomial populations, Sankhy¯ a: the indian journal of statistics (1946) 401–406

    A. Bhattacharyya, On a measure of divergence between tw o multinomial populations, Sankhy¯ a: the indian journal of statistics (1946) 401–406

    1946

  44. [44]

    R. Kashyap, The perfect marriage and much more: Combini ng dimension re- duction, distance measures and covariance, Physica A: Statistical Mechanics and its Applications 536 (2019) 120938

    2019

  45. [45]

    K. He, X. Zhang, S. Ren, J. Sun, Deep residual learning fo r image recogni- tion, in: Proceedings of the IEEE conference on computer vis ion and pattern recognition, 2016, pp. 770–778

    2016

  46. [46]

    D. P . Kingma, T. Salimans, M. Welling, V ariational drop out and the local reparameterization trick, Advances in neural information processing systems 28 (2015)

    2015

  47. [47]

    A. Roy, A. Shah, K. Shah, P . Dhar, A. Cherian, R. Chellapp a, Felmi: few shot learning with hard mixup, Advances in Neural Informati on Processing Systems 35 (2022) 24474–24486

    2022

  48. [48]

    J. Deng, W. Dong, R. Socher, L.-J. Li, K. Li, L. Fei-Fei, I magenet: A large- scale hierarchical image database, in: 2009 IEEE conferenc e on computer vision and pattern recognition, Ieee, 2009, pp. 248–255

    2009

  49. [49]

    K. Lee, S. Maji, A. Ravichandran, S. Soatto, Meta-learn ing with differen- tiable convex optimization, in: Proceedings of the IEEE/CV F conference on computer vision and pattern recognition, 2019, pp. 10657–1 0665

    2019

  50. [50]

    Simon, P

    C. Simon, P . Koniusz, R. Nock, M. Harandi, Adaptive subspaces for few-shot learning, in: Proceedings of the IEEE/CVF conference on com puter vision and pattern recognition, 2020, pp. 4136–4145

    2020

  51. [51]

    Y . Tian, Y . Wang, D. Krishnan, J. B. Tenenbaum, P . Isola, Rethinking few- shot image classification: a good embedding is all you need?, in: Computer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, A ugust 23– 28, 2020, Proceedings, Part XIV 16, Springer, 2020, pp. 266– 282. 50

    2020

  52. [52]

    M. N. Rizve, S. Khan, F. S. Khan, M. Shah, Exploring compl ementary strengths of invariant and equivariant representations fo r few-shot learning, in: Proceedings of the IEEE/CVF conference on computer visi on and pattern recognition, 2021, pp. 10836–10846

    2021

  53. [53]

    J. Ma, H. Xie, G. Han, S.-F. Chang, A. Galstyan, W. Abd-Almageed, Partner- assisted learning for few-shot image classification, in: Pr oceedings of the IEEE/CVF International Conference on Computer Vision, 202 1, pp. 10573– 10582

  54. [54]

    Rajasegaran, S

    J. Rajasegaran, S. Khan, M. Hayat, F. S. Khan, M. Shah, Se lf- supervised knowledge distillation for few-shot learning, arXiv preprint arXiv:2006.09785 (2020)

    Pith/arXiv arXiv 2006

  55. [55]

    Y . Jian, L. Torresani, Label hallucination for few-sho t classification, in: Pro- ceedings of the AAAI Conference on Artificial Intelligence, V ol. 36, 2022, pp. 7005–7014

    2022

  56. [56]

    G. Y . Lee, T. Dam, D. P . Poenar, V . N. Duong, M. M. Ferdaus, Hela-vfa: A hellinger distance-attention-based feature aggregatio n network for few- shot classification, in: Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, 2024, pp. 2173–2183

    2024

  57. [57]

    Oreshkin, P

    B. Oreshkin, P . Rodr´ ıguez L´ opez, A. Lacoste, Tadam: Task dependent adap- tive metric for improved few-shot learning, Advances in neu ral information processing systems 31 (2018)

    2018

  58. [58]

    Q. Sun, Y . Liu, T.-S. Chua, B. Schiele, Meta-transfer le arning for few-shot learning, in: Proceedings of the IEEE/CVF conference on com puter vision and pattern recognition, 2019, pp. 403–412

    2019

  59. [59]

    Zhang, Y

    C. Zhang, Y . Cai, G. Lin, C. Shen, Deepemd: Few-shot imag e classifica- tion with differentiable earth mover’s distance and struct ured classifiers, in: Proceedings of the IEEE/CVF conference on computer vision a nd pattern recognition, 2020, pp. 12203–12213

    2020

  60. [60]

    Afrasiyabi, J.-F

    A. Afrasiyabi, J.-F. Lalonde, C. Gagn´ e, Associative a lignment for few- shot image classification, in: Computer Vision–ECCV 2020: 1 6th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Part V 16, Springer, 2020, pp. 18–35. 51

    2020

  61. [61]

    Y . Gao, N. Fei, G. Liu, Z. Lu, T. Xiang, Contrastive proto type learning with augmented embeddings for few-shot learning, in: Uncertain ty in Artificial Intelligence, PMLR, 2021, pp. 140–150

    2021

  62. [62]

    Ravichandran, R

    A. Ravichandran, R. Bhotika, S. Soatto, Few-shot learn ing with embedded class models and shot-free meta training, in: Proceedings o f the IEEE/CVF international conference on computer vision, 2019, pp. 331 –339

    2019

  63. [63]

    H.-J. Y e, H. Hu, D.-C. Zhan, F. Sha, Few-shot learning vi a embedding adap- tation with set-to-set functions, in: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2020, pp. 8808– 8817

    2020

  64. [64]

    B. Liu, Y . Cao, Y . Lin, Q. Li, Z. Zhang, M. Long, H. Hu, Nega tive mar- gin matters: Understanding margin in few-shot classificati on, in: Computer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, A ugust 23– 28, 2020, Proceedings, Part IV 16, Springer, 2020, pp. 438–4 55

    2020

  65. [65]

    Z. Shen, Z. Liu, J. Qin, M. Savvides, K.-T. Cheng, Partia l is better than all: revisiting fine-tuning strategy for few-shot learning, in: Proceedings of the AAAI Conference on Artificial Intelligence, V ol. 35, 2021, pp. 9594–9602

    2021

  66. [66]

    N. Fei, Z. Lu, T. Xiang, S. Huang, Melr: Meta-learning vi a modeling episode-level relationships for few-shot learning, in: In ternational Confer- ence on Learning Representations, 2020

    2020

  67. [67]

    S. W. Y oon, J. Seo, J. Moon, Tapnet: Neural network augme nted with task- adaptive projection for few-shot learning, in: Internatio nal conference on machine learning, PMLR, 2019, pp. 7115–7123

    2019

  68. [68]

    Zhang, J

    M. Zhang, J. Zhang, Z. Lu, T. Xiang, M. Ding, S. Huang, Iep t: Instance- level and episode-level pretext tasks for few-shot learnin g, in: International Conference on Learning Representations, 2020

    2020

  69. [69]

    Tolstikhin, O

    I. Tolstikhin, O. Bousquet, S. Gelly, B. Schoelkopf, Wa sserstein auto- encoders, arXiv preprint arXiv:1711.01558 (2017)

    Pith/arXiv arXiv 2017

  70. [70]

    Z. Chen, P . Liu, Towards efficient variational auto-enc oder using wasser- stein distance, in: 2022 IEEE International Conference on I mage Processing (ICIP), IEEE, 2022, pp. 81–85. 52

    2022

  71. [71]

    Heusel, H

    M. Heusel, H. Ramsauer, T. Unterthiner, B. Nessler, S. H ochreiter, Gans trained by a two time-scale update rule converge to a local na sh equilibrium, Advances in neural information processing systems 30 (2017 )

    2017

  72. [72]

    Szegedy, V

    C. Szegedy, V . V anhoucke, S. Ioffe, J. Shlens, Z. Wojna, Rethinking the in- ception architecture for computer vision, in: Proceedings of the IEEE con- ference on computer vision and pattern recognition, 2016, p p. 2818–2826

    2016

  73. [73]

    G. Y . Lee, T. Dam, M. M. Ferdaus, D. P . Poenar, V . N. Duong,Watt-effnet: A lightweight and accurate model for classifying aerial disa ster images, IEEE Geoscience and Remote Sensing Letters 20 (2023) 1–5

    2023

  74. [74]

    Sandler, A

    M. Sandler, A. Howard, M. Zhu, A. Zhmoginov, L.-C. Chen, Mobilenetv2: Inverted residuals and linear bottlenecks, in: Proceeding s of the IEEE con- ference on computer vision and pattern recognition, 2018, p p. 4510–4520

    2018

  75. [75]

    Y an, Feature aligning few shot learning method using local descriptors weighted rules, arXiv preprint arXiv:2408.14192 (2024)

    B. Y an, Feature aligning few shot learning method using local descriptors weighted rules, arXiv preprint arXiv:2408.14192 (2024)

    Pith/arXiv arXiv 2024

  76. [76]

    Boudiaf, I

    M. Boudiaf, I. Ziko, J. Rony, J. Dolz, P . Piantanida, I. B en Ayed, Informa- tion maximization for few-shot learning, Advances in Neura l Information Processing Systems 33 (2020) 2445–2457

    2020

  77. [77]

    B. Wang, L. Li, M. V erma, Y . Nakashima, R. Kawasaki, H. Na gahara, Mtunet: Few-shot image classification with visual explanations, in: Proceed- ings of the IEEE/CVF Conference on Computer Vision and Patte rn Recog- nition, 2021, pp. 2294–2298. 53

    2021

This paper was first reviewed by deepseek-v4-flash on August 4, 2026.