Pith. sign in

REVIEW 4 major objections 7 minor 63 references

Embedding Space Allocation with Angle-Norm Joint Classifiers for Few-Shot Class-Incremental Learning

T0 review · 4 major / 7 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 SAAN claims a plug-in class-center space allocation and angle-norm classifier that raises final-session accuracy by over 3% in few-shot class-incremental learning.

desk verdict A clean plug-in method with credible gains on two benchmarks, but the advertised SOTA and 3% claims are overstated and the main tables lack error bars. read the letter →

arxiv 2411.09250 v2 pith:TN4OVWFG submitted 2024-11-14 cs.CV

Dunwei Tu , Huiyu Yi , Tieyi Zhang , Ruotong Li , Furao Shen , Jian Zhao This is my paper · ORCID
classification cs.CV
keywords few-shotclass-incrementallearningembeddingspaceallocationcosinecenterlossangle-normjointclassifiernormdistributionprototypeplug-inmethodnearestclassmean
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 tries to solve few-shot class-incremental learning (FSCIL), where a model must keep recognizing old classes while learning new ones from only a handful of examples. It proposes SAAN, a two-part scheme: CCSA partitions the embedding space by assigning each session a reserved subspace through orthogonal class centers and a cosine center loss, and ANJ replaces the cosine-only nearest-class-mean classifier with one that also models each class's embedding-norm distribution and combines angle and norm scores. The paper's central claim is that SAAN is competitive by itself and, more importantly, can be plugged into existing methods such as FACT and SAVC, raising final-session accuracy by over 3% on CIFAR100, CUB200, and miniImageNet while reducing forgetting. If that holds, existing feature-freezing FSCIL methods can be upgraded by adding SAAN's allocation loss and classifier, gaining accuracy on the final session without changing the frozen-feature paradigm.

What carries the argument

The machinery is the pair (CCSA, ANJ). CCSA is the space-allocating component: it fixes a set of orthogonal class centers, assigns them to classes via the Hungarian algorithm, and trains with a cosine center loss so that sample embeddings gather around their assigned center and away from others, with momentum center updates that decay so the reserved spaces survive. ANJ is the classifier component: it computes class means from normalized embeddings only in the base session (2SNCM), fits log-norm distributions to embedding lengths, and produces joint angle-norm logits by compressing the norm probability and multiplying it with the cosine logit. The argument depends on two mechanisms holding at once: new classes actually land in the reserved subspaces, and norm differences carry label information that the cosine score misses.

What would settle it

Compute a Q-Q plot of feature-vector log lengths on the last incremental session of a new dataset, then compare SAVC with and without the norm-based score; if a visible deviation from the straight line coincides with the norm term adding no final-session gain, the normality assumption is the part that fails.

Watch

Extended reading notes

Core claim

The central discovery, on the paper's own terms, is that two weaknesses of FSCIL—old classes sprawling over the whole embedding space and nearest-class-mean classifiers ignoring vector length—can be addressed together. CCSA pre-generates mutually orthogonal class centers, matches them to classes with the Hungarian algorithm using cosine distance, and trains with a cosine center loss $L_{cc}=\alpha L_1+\beta L_2$ that pulls samples toward their own center and away from others, so each incremental session gets a reserved subspace. ANJ observes that embedding log norms differ between classes and sessions, models them as normal distributions (one per base class, one shared normal for all incremental classes), and forms joint logits $z_j = z^1_j (z^2_j)^C$, multiplying cosine similarity by a compressed norm probability. The experimental case is that SAAN alone beats FACT on all three benchmarks and that FACT+SAAN and SAVC+SAAN improve final-session accuracy by over 3% and lower forgetting.

Load-bearing premise

The load-bearing premise is that the logarithms of feature-vector lengths follow a normal distribution—each base class with its own mean and variance, and all incremental classes sharing one mean and variance—so if the actual distribution is not normal on a new dataset, the norm-based scores become miscalibrated and the plug-in gains shrink.

Editorial extensions

If this is right

  • Plugging SAAN into FACT and SAVC lifts final-session accuracy by over 3% on CIFAR100, CUB200, and miniImageNet and reduces average forgetting by about 2.6 percentage points.
  • Used alone, SAAN beats FACT on all three standard benchmarks and beats SAVC on CIFAR100, showing the space allocation and norm classifier are competitive without contrastive pretraining.
  • The plug-in gains carry over to open-ended, imbalanced long-horizon settings: on CompCars with 20 incremental sessions, SAVC+SAAN exceeds SAVC by 2.83% in last-session accuracy.
  • Ablations show the largest single contribution comes from the cosine center pull that reserves space, with the norm-distribution term adding further gains, so both halves of the framework are claimed to be necessary for the full improvement.

Reading between the lines

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

  • Because ANJ only changes how a frozen embedding is scored, the norm-logit stage could be tested as a generic upgrade to other nearest-class-mean pipelines beyond FSCIL, provided the log-norm normality assumption is verified first.
  • The shared normal distribution for all incremental classes is a strong simplification; as an incremental run grows longer, fitting per-class or per-session norm distributions could recover information the shared prior throws away, and the paper's own session-8 Q-Q deviation points to where that limit would show up.
  • The Hungarian center assignment fixes the allocation once; a natural extension is to test whether allowing centers to be re-allocated when class semantics shift, or using hierarchical centers for very large class sets, changes the plug-in gain.
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 / 7 minor

Summary. The paper proposes SAAN, a plug-in framework for few-shot class-incremental learning, consisting of two components: CCSA, which partitions the embedding space by assigning orthogonal class centers via the Hungarian algorithm and guiding embeddings with a cosine center loss, and ANJ, which models the log-norm distribution of each class and combines angle-based NCM logits with norm-based logits. The authors claim that SAAN achieves state-of-the-art performance and that adding it as a plug-in to FACT or SAVC improves final-session accuracy by over 3% across CIFAR100, CUB200, and miniImageNet. Experiments also cover an open-ended imbalanced CompCars setting, ablations, hyperparameter sensitivity, and a computational cost analysis.

Significance. If the plug-in property were reliably established, this would be a practically useful contribution: the method is simple, adds negligible parameters and inference cost (Appendix D), and is designed to integrate with existing virtual-class FSCIL methods. The paper provides useful ablations showing the individual contributions of space allocation, the two-stage NCM, and norm distribution, and it includes some repeated experiments in the open-ended setting and for 2SNCM. However, the central quantitative claim is currently overstated and the main tables do not provide enough statistical evidence to support the claimed consistency of the plug-in gains.

major comments (4)
  1. [Abstract and Contribution 3, Tables 1-3] The central claim that SAAN 'improves the final round accuracy by over 3% across three datasets and two methods' is not supported by the paper's own tables. In Table 3, FACT+SAAN improves from 51.93 to 52.11 on CIFAR100, which is +0.18 percentage points, and SAVC+SAAN improves from 51.64 to 54.22, which is +2.58 percentage points. Only the CUB200 and miniImageNet SAVC+SAAN comparisons exceed 3%. In addition, the abstract's 'state-of-the-art' statement is too strong: standalone SAAN is below NC-FSCIL and M2SD on all three datasets and below SAVC on CUB200 and miniImageNet. The claim should be corrected to report the actual per-dataset gains, or additional experiments should be provided if the 3% claim is meant to be an average or a best-case result.
  2. [Tables 1-3 and Section 5.1] The main plug-in comparisons are single runs reported without error bars, seeds, or confidence intervals. FSCIL results depend on random initialization and on the order of incremental sessions, so paired differences of +0.18 and +2.58 percentage points may be within run-to-run noise. The paper already uses repeated experiments in Section 5.3 and Appendix E, so the same practice should be applied to the main comparisons: report the mean and standard deviation over at least three to five seeds for FACT, SAVC, FACT+SAAN, and SAVC+SAAN on all three datasets. Without this, the claim that SAAN 'consistently' improves both methods on all datasets is not statistically established.
  3. [Section 5.5, Figures 7 and 8] The hyperparameters alpha=2, beta=0.4, and C=0.005 are selected by grid search on the final-session accuracy of the same test benchmarks that are then used to produce the main results in Tables 1-3. This is a form of test-set tuning, and it weakens the claim that the plug-in gains reflect a general property of SAAN rather than per-dataset hyperparameter fitting. The authors should either use a held-out validation split for hyperparameter selection or show that a fixed, non-tuned hyperparameter setting yields qualitatively similar improvements across all datasets.
  4. [Section 4.2.2, Equation 10, Figure 10] The ANJ classifier relies on the assumption that log-norms are normally distributed for each base class and that all incremental classes share a single normal distribution. The evidence is limited to histograms and one Q-Q plot, and the authors themselves note that session 8 deviates from normality. Since this distributional assumption is a core component of ANJ, the paper should provide a quantitative goodness-of-fit test or a sensitivity analysis showing that the plug-in gains are not an artifact of the normality assumption. For example, the authors could compare the proposed parametric model against a nonparametric norm distribution or evaluate calibration of the norm logits.
minor comments (7)
  1. [Tables 1-3] M2SD is listed as a comparison method in Section 5.2 but is missing from Table 3 (CIFAR100); if results are unavailable, the paper should state this explicitly.
  2. [Section 6] There is a typo: 'SSAN' appears in the conclusion where 'SAAN' is intended.
  3. [Section 5.2] The sentence 'which allows SSAN to achieve better performance' contains the same typo; it should read 'SAAN'.
  4. [Table 4] The table header repeats the same description for both 'Delta last' and 'Delta average'; the second should refer to average accuracy rather than 'last sessions'.
  5. [Table 6] The distribution name 'Patero' is a typo and should be 'Pareto', consistent with the text.
  6. [Section 2] In the related work text, 'MS2D' appears to be a typo for 'M2SD'.
  7. [Figure 6] The harmonic mean is mentioned in the text but is not defined; please specify how it is computed and why it is used.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity; SAAN's components are independently defined training objectives/classifiers and the plug-in claim is an external benchmark result, not a derivation from its own inputs.

full rationale

SAAN's two components are defined independently of the accuracy numbers they are claimed to produce. CCSA is a cosine center loss (Eqs. 3-5) pulling embeddings to Hungarian-assigned orthogonal centers; ANJ builds per-class/shared log-normal models from training embeddings (Sec. 4.2.2) and combines angle and norm logits via Eq. 10 and z_j = z1_j (z2_j)^C. The final-session accuracies in Tables 1-4 are measured on external benchmarks (CIFAR100, CUB200, miniImageNet, CompCars) against independent baselines, not derived from the loss definitions or from a fitted parameter renamed as the target. The normality assumption is empirically checked with Q-Q plots (Fig. 10) and distribution ablations (Table 6), and the authors explicitly note the session-8 long-tail deviation. The plug-in claim is an empirical comparison; the absence of error bars, single runs, and grid-search selection of alpha, beta, and C on the same benchmarks are threats to statistical validity, not circularity. No load-bearing self-citation occurs: the only same-group citations ([20,21]) support data augmentation in related work, and no uniqueness theorem or prior-work premise is imported to force the method. Therefore no circular step satisfying the evidentiary standard can be identified, and the appropriate finding is no significant circularity.

Assumptions & free parameters 6 free parameters · 5 assumptions · 0 invented entities

No new physical or ontological entities are introduced; class centers are ordinary learnable parameters. The main load-bearing assumptions are the log-normality of embedding norms, the shared incremental distribution, the feasibility of d orthogonal centers, and the semantic meaning of cosine distances between centers and class means.

free parameters (6)
  • alpha (loss weight for L1) = 2 (CUB200; alpha/beta ratio 5 across datasets)
    Grid search on benchmark last-session accuracy; no held-out validation described.
  • beta (loss weight for L2) = 0.4 at alpha=2
    Grid search on benchmark last-session accuracy; the paper notes that too large a beta weakens the pull toward the correct center.
  • C (compression coefficient for norm logits) = 0.005 for all three datasets
    Optimal in grid search; larger values make norm logits overly dominant and hurt accuracy.
  • eta (center moving rate) = positive values in 0.25 to 2; exact value not reported
    Chosen by grid search; eta=0 gives fixed centers, eta with no decay gives continuous updating that fails to allocate space.
  • lambda (center moving decay rate) = 0.1
    Grid search; performance is relatively insensitive across a range.
  • Mu and sigma for log-norm distributions = Sample estimates from training embeddings; per base class and shared across incremental sessions
    Used in Eq. 10 to compute norm logits; these are fitted from data, not derived.
assumptions (5)
  • domain assumption Log-norm of embeddings within a class is normally distributed, and incremental classes share one normal distribution.
    Invoked in Sec. 4.2.2 to define norm logits; supported only by histograms and one Q-Q plot (Fig. 10), which the authors note deviates in session 8.
  • domain assumption The embedding dimension d is large enough to host d mutually orthogonal class centers, and the total number of classes never exceeds d.
    Sec. 4.1.1 sets d orthogonal centers as the maximum possible in d-dimensional space; if future classes exceed d, orthogonal allocation is impossible.
  • domain assumption Cosine distance between class centers and class means reflects semantic similarity, so Hungarian assignment gives semantically coherent allocation.
    Sec. 4.1.1 and 4.1.3; this is a heuristic supported only by qualitative t-SNE visualization, not by a formal measure.
  • domain assumption Norm differences between base and incremental classes are a stable signal usable for classification rather than a dataset artifact.
    Sec. 4.2.1 observes norm differences in Fig. 4 and exploits them; the shared incremental distribution means norms mainly separate sessions, not classes.
  • standard math Hungarian algorithm solves the assignment problem optimally.
    Used in Sec. 4.1.1 to assign class centers; this is a standard result and is not in question.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Embedding Space Allocation with Angle-Norm Joint Classifiers for Few-Shot Class-Incremental Learning." pith.science (2026). https://pith.science/paper/TN4OVWFG

@misc{pith2026241109250,
  author       = {Pith},
  title        = {Pith review of: Embedding Space Allocation with Angle-Norm Joint Classifiers for Few-Shot Class-Incremental Learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/TN4OVWFG}},
  note         = {Machine review of arXiv:2411.09250}
}
read the original abstract

Few-shot class-incremental learning (FSCIL) aims to continually learn new classes from only a few samples without forgetting previous ones, requiring intelligent agents to adapt to dynamic environments. FSCIL combines the characteristics and challenges of class-incremental learning and few-shot learning: (i) Current classes occupy the entire feature space, which is detrimental to learning new classes. (ii) The small number of samples in incremental rounds is insufficient for fully training. In existing mainstream virtual class methods, for addressing the challenge (i), they attempt to use virtual classes as placeholders. However, new classes may not necessarily align with the virtual classes. For the challenge (ii), they replace trainable fully connected layers with Nearest Class Mean (NCM) classifiers based on cosine similarity, but NCM classifiers do not account for sample imbalance issues. To address these issues in previous methods, we propose the class-center guided embedding Space Allocation with Angle-Norm joint classifiers (SAAN) learning framework, which provides balanced space for all classes and leverages norm differences caused by sample imbalance to enhance classification criteria. Specifically, for challenge (i), SAAN divides the feature space into multiple subspaces and allocates a dedicated subspace for each session by guiding samples with the pre-set category centers. For challenge (ii), SAAN establishes a norm distribution for each class and generates angle-norm joint logits. Experiments demonstrate that SAAN can achieve state-of-the-art performance and it can be directly embedded into other SOTA methods as a plug-in, further enhancing their performance.

Figures

Figures reproduced from arXiv: 2411.09250 by the authors.

Figure 1
Figure 1. The motivation of SAAN. In only cross-entropy supervision, old categories [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Illustration of CCSA. CCSA sets orthogonal class centers and assigns these cen [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗
Figure 3
Figure 3. Overall pipeline of ANJ. ANJ consists 2SNCM and Norm Distribution. 2SNCM [PITH_FULL_IMAGE:figures/full_fig_p013_3.png] view at source ↗
Figures from the paper (7 more)
Figure 4
Figure 4. Figure 4: Histogram of the distribution of the log norms of sample embedding on [PITH_FULL_IMAGE:figures/full_fig_p014_4.png]
Figure 5
Figure 5. Figure 5: The t-SNE visualization on CIFAR100 dataset of the embeddings. Classes 0- [PITH_FULL_IMAGE:figures/full_fig_p024_5.png]
Figure 6
Figure 6. Figure 6: Accuracy of base and novel classes at the last session. [PITH_FULL_IMAGE:figures/full_fig_p024_6.png]
Figure 7
Figure 7. Figure 7: The parameter grid search for the last session accuracy of SAVC + SAAN on [PITH_FULL_IMAGE:figures/full_fig_p025_7.png]
Figure 8
Figure 8. Figure 8: Last session accuracy of SAVC + SAAN with different hyper-parameters on [PITH_FULL_IMAGE:figures/full_fig_p026_8.png]
Figure 9
Figure 9. Figure 9: The angle distributions between the sample embeddings and their corresponding [PITH_FULL_IMAGE:figures/full_fig_p027_9.png]
Figure 10
Figure 10. Figure 10: Quantile-Quantile plot of embedding log norms on [PITH_FULL_IMAGE:figures/full_fig_p027_10.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

63 extracted references · 50 canonical work pages

  1. [1]

    J. Deng, W. Dong, R. Socher, L.-J. Li, K. Li, L. Fei-Fei, Imagenet: A large-scale hierarchical image database, in: CVPR, 2009, pp. 248–255. doi:10.1109/CVPR.2009.5206848

    arXiv 2009

  2. [2]

    K. He, X. Zhang, S. Ren, J. Sun, Deep residual learning for image recognition, in: CVPR, 2016, pp. 770–778

    work page 2016

  3. [3]

    Simonyan, A

    K. Simonyan, A. Zisserman, Very deep convolutional networks for large- scale image recognition, arXiv preprint arXiv:1409.1556 (2014)

    arXiv 2014

  4. [4]

    M. Tan, R. Pang, Q. V. Le, Efficientdet: Scalable and efficient object detection, in: CVPR, 2020, pp. 10781–10790

    work page 2020

  5. [5]

    Y. Wang, Q. Yao, J. T. Kwok, L. M. Ni, Generalizing from a few examples: A survey on few-shot learning, CSUR 53 (2020) 1–34

    work page 2020

  6. [6]

    F. Sung, Y. Yang, L. Zhang, T. Xiang, P. H. Torr, T. M. Hospedales, Learning to compare: Relation network for few-shot learning, in: CVPR, 2018, pp. 1199–1208

    work page 2018

  7. [7]

    Snell, K

    J. Snell, K. Swersky, R. Zemel, Prototypical networks for few-shot learn- ing, in: NeurIPS, volume 30, 2017

    work page 2017

  8. [8]

    X. Tao, X. Hong, X. Chang, S. Dong, X. Wei, Y. Gong, Few-shot class- incremental learning, in: CVPR, 2020, pp. 12183–12192

    work page 2020

Show all 63 references
  1. [9]

    Kalla, S

    J. Kalla, S. Biswas, S3c: Self-supervised stochastic classifiers for few- shot class-incremental learning, in: ECCV, Springer, 2022, pp. 432–448. 33

    2022

  2. [10]

    Zhang, N

    C. Zhang, N. Song, G. Lin, Y. Zheng, P. Pan, Y. Xu, Few-shot incre- mental learning with continually evolved classifiers, in: CVPR, 2021, pp. 12455–12464

    2021

  3. [11]

    D. Zhou, F. Wang, H. Ye, L. Ma, S. Pu, D. Zhan, Forward compatible few-shot class-incremental learning, in: CVPR, 2022, pp. 9046–9056

    2022

  4. [12]

    Y. Yang, H. Yuan, X. Li, Z. Lin, P. Torr, D. Tao, Neural collapse inspired feature-classifier alignment for few-shot class incremental learning, arXiv preprint arXiv:2302.03004 (2023)

    2023 arXiv

  5. [13]

    J. Lin, Z. Wu, W. Lin, J. Huang, R. Luo, M2sd: Multiple mixing self- distillation for few-shot class-incremental learning, in: AAAI, volume 38, 2024, pp. 3422–3431

    2024

  6. [14]

    Cheraghian, S

    A. Cheraghian, S. Rahman, P. Fang, S. K. Roy, L. Petersson, M. Harandi, Semantic-aware knowledge distillation for few-shot class- incremental learning, in: CVPR, 2021, pp. 2534–2543

    2021

  7. [15]

    S. Dong, X. Hong, X. Tao, X. Chang, X. Wei, Y. Gong, Few-shot class-incremental learning via relation knowledge distillation, in: AAAI, volume 35, 2021, pp. 1255–1263

    2021

  8. [16]

    Z. Song, Y. Zhao, Y. Shi, P. Peng, L. Yuan, Y. Tian, Learning with fan- tasy: Semantic-aware virtual contrastive constraint for few-shot class- incremental learning, in: CVPR, 2023, pp. 24183–24192

    2023

  9. [17]

    Mensink, J

    T. Mensink, J. Verbeek, F. Perronnin, G. Csurka, Distance-based image classification: Generalizing to new classes at near-zero cost, TPAMI 35 (2013) 2624–2637

    2013

  10. [18]

    Kothapalli, Neural collapse: A review on modelling principles and generalization, arXiv preprint arXiv:2206.04041 (2022)

    V. Kothapalli, Neural collapse: A review on modelling principles and generalization, arXiv preprint arXiv:2206.04041 (2022)

    2022 arXiv

  11. [19]

    X. Han, V. Papyan, D. L. Donoho, Neural collapse under mse loss: Proximity to and dynamics on the central path, arXiv preprint arXiv:2106.02073 (2021)

    2021 arXiv

  12. [20]

    S. Yang, F. Shen, J. Zhao, Entaugment: Entropy-driven adaptive data augmentation framework for image classification, in: ECCV, Springer, 2024, pp. 197–214. 34

    2024

  13. [21]

    S. Yang, S. Guo, J. Zhao, F. Shen, Investigating the effectiveness of data augmentation from similarity and diversity: An empirical study, Pattern Recognition 148 (2024) 110204

    2024

  14. [22]

    Guerriero, B

    S. Guerriero, B. Caputo, T. Mensink, Deepncm: Deep nearest class mean classifiers, 2018

    2018

  15. [23]

    Thongtan, T

    T. Thongtan, T. Phienthrakul, Sentiment classification using document embeddings trained with cosine similarity, in: ACL: Student Research Workshop, 2019, pp. 407–414

    2019

  16. [24]

    K. Zhou, K. Ethayarajh, D. Card, D. Jurafsky, Problems with cosine as a measure of embedding similarity for high frequency words, arXiv preprint arXiv:2205.05092 (2022)

    2022 arXiv

  17. [25]

    Ahmed, A

    N. Ahmed, A. Kukleva, B. Schiele, Orco: Towards better generalization via orthogonality and contrast for few-shot class-incremental learning, in: CVPR, 2024, pp. 28762–28771

    2024

  18. [26]

    Y. Fu, Y. Fu, Y.-G. Jiang, Meta-fdmixup: Cross-domain few-shot learn- ing guided by labeled target data, in: MM, 2021, pp. 5326–5334

    2021

  19. [27]

    S. Ravi, H. Larochelle, Optimization as a model for few-shot learning, in: ICLR, 2016

    2016

  20. [28]

    D. Wang, Y. Cheng, M. Yu, X. Guo, T. Zhang, A hybrid approach with optimization-based and metric-based meta-learner for few-shot learning, Neurocomputing 349 (2019) 202–211

    2019

  21. [29]

    Arnold, S

    S. Arnold, S. Iqbal, F. Sha, When maml can adapt fast and how to assist when it cannot, in: AISTATS, PMLR, 2021, pp. 244–252

    2021

  22. [30]

    Nichol, J

    A. Nichol, J. Achiam, J. Schulman, On first-order meta-learning algo- rithms, arXiv preprint arXiv:1803.02999 (2018)

    2018 arXiv

  23. [31]

    Oreshkin, P

    B. Oreshkin, P. Rodr ´ ıguez L´ opez, A. Lacoste, Tadam: Task dependent adaptive metric for improved few-shot learning, in: NeurIPS, volume 31, 2018

    2018

  24. [32]

    B. Liu, Y. Cao, Y. Lin, Q. Li, Z. Zhang, M. Long, H. Hu, Negative margin matters: Understanding margin in few-shot classification, in: ECCV, Springer, 2020, pp. 438–455. 35

    2020

  25. [33]

    Zhang, Y

    C. Zhang, Y. Cai, G. Lin, C. Shen, Deepemd: Few-shot image classifica- tion with differentiable earth mover’s distance and structured classifiers, in: CVPR, 2020, pp. 12203–12213

    2020

  26. [34]

    Triantafillou, R

    E. Triantafillou, R. Zemel, R. Urtasun, Few-shot learning through an information retrieval lens, in: NeurIPS, volume 30, 2017

    2017

  27. [35]

    D. Wang, M. Zhang, Y. Xu, W. Lu, J. Yang, T. Zhang, Metric-based meta-learning model for few-shot fault diagnosis under multiple limited data conditions, MSSP 155 (2021) 107510

    2021

  28. [36]

    Mittal, S

    S. Mittal, S. Galesso, T. Brox, Essentials for class incremental learning, in: CVPR, 2021, pp. 3513–3522

    2021

  29. [37]

    Masana, X

    M. Masana, X. Liu, B. Twardowski, M. Menta, A. D. Bagdanov, J. Van De Weijer, Class-incremental learning: survey and performance evalua- tion on image classification, TPAMI 45 (2022) 5513–5533

    2022

  30. [38]

    Zhang, J

    J. Zhang, J. Zhang, S. Ghosh, D. Li, S. Tasci, L. Heck, H. Zhang, C.- C. J. Kuo, Class-incremental learning via deep model consolidation, in: W ACV, 2020, pp. 1131–1140

    2020

  31. [39]

    Z. Wang, Z. Zhang, C.-Y. Lee, H. Zhang, R. Sun, X. Ren, G. Su, V. Perot, J. Dy, T. Pfister, Learning to prompt for continual learn- ing, in: CVPR, 2022, pp. 139–149

    2022

  32. [40]

    S. Yan, J. Xie, X. He, Der: Dynamically expandable representation for class incremental learning, in: CVPR, 2021, pp. 3014–3023

    2021

  33. [41]

    Rebuffi, A

    S.-A. Rebuffi, A. Kolesnikov, G. Sperl, C. H. Lampert, icarl: Incremen- tal classifier and representation learning, in: CVPR, 2017, pp. 2001– 2010

    2017

  34. [42]

    B. Zhao, X. Xiao, G. Gan, B. Zhang, S.-T. Xia, Maintaining discrimi- nation and fairness in class incremental learning, in: CVPR, 2020, pp. 13208–13217

    2020

  35. [43]

    Y. Wu, Y. Chen, L. Wang, Y. Ye, Z. Liu, Y. Guo, Y. Fu, Large scale incremental learning, in: CVPR, 2019, pp. 374–382. 36

    2019

  36. [44]

    Rajasegaran, S

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

    2020 arXiv

  37. [45]

    Z. Li, D. Hoiem, Learning without forgetting, TPAMI 40 (2017) 2935– 2947

    2017

  38. [46]

    Hinton, O

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

    2015 arXiv

  39. [47]

    Aljundi, F

    R. Aljundi, F. Babiloni, M. Elhoseiny, M. Rohrbach, T. Tuytelaars, Memory aware synapses: Learning what (not) to forget, in: ECCV, 2018, pp. 139–154

    2018

  40. [48]

    X. Liu, M. Masana, L. Herranz, J. Van de Weijer, A. M. Lopez, A. D. Bagdanov, Rotate your networks: Better weight consolidation and less catastrophic forgetting, in: ICPR, IEEE, 2018, pp. 2262–2268

    2018

  41. [49]

    Zenke, B

    F. Zenke, B. Poole, S. Ganguli, Continual learning through synaptic intelligence, in: ICML, PMLR, 2017, pp. 3987–3995

    2017

  42. [50]

    Krizhevsky, G

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

    2009

  43. [51]

    Y. Wen, K. Zhang, Z. Li, Y. Qiao, A discriminative feature learning approach for deep face recognition, in: ECCV, Springer, 2016, pp. 499– 515

    2016

  44. [52]

    H. W. Kuhn, The hungarian method for the assignment problem, Naval research logistics quarterly 2 (1955) 83–97

    1955

  45. [53]

    Munkres, Algorithms for the assignment and transportation problems, Journal of the society for industrial and applied mathematics 5 (1957) 32–38

    J. Munkres, Algorithms for the assignment and transportation problems, Journal of the society for industrial and applied mathematics 5 (1957) 32–38

    1957

  46. [54]

    Goldberg, O

    Y. Goldberg, O. Levy, word2vec explained: deriving mikolov et al.’s negative-sampling word-embedding method, arXiv preprint arXiv:1402.3722 (2014)

    2014 arXiv

  47. [55]

    Mikolov, I

    T. Mikolov, I. Sutskever, K. Chen, G. S. Corrado, J. Dean, Distributed representations of words and phrases and their compositionality, NIPS 26 (2013). 37

    2013

  48. [56]

    C. Wah, S. Branson, P. Welinder, P. Perona, S. Belongie, The caltech- ucsd birds-200-2011 dataset (2011)

    2011

  49. [57]

    Russakovsky, J

    O. Russakovsky, J. Deng, H. Su, J. Krause, S. Satheesh, S. Ma, Z. Huang, A. Karpathy, A. Khosla, M. Bernstein, et al., Imagenet large scale visual recognition challenge, IJCV 115 (2015) 211–252

    2015

  50. [58]

    L. Yang, P. Luo, C. Change Loy, X. Tang, A large-scale car dataset for fine-grained categorization and verification, in: CVPR, 2015, pp. 3973–3981

    2015

  51. [59]

    F. M. Castro, M. J. Mar ´ ın-Jim´ enez, N. Guil, C. Schmid, K. Alahari, End-to-end incremental learning, in: ECCV, 2018, pp. 233–248

    2018

  52. [60]

    Vinyals, C

    O. Vinyals, C. Blundell, T. Lillicrap, k. kavukcuoglu, D. Wierstra, Matching networks for one shot learning, in: D. Lee, M. Sugiyama, U. Luxburg, I. Guyon, R. Garnett (Eds.), NeurIPS, volume 29, 2016

    2016

  53. [61]

    K. He, H. Fan, Y. Wu, S. Xie, R. Girshick, Momentum contrast for unsupervised visual representation learning, in: CVPR, 2020, pp. 9729– 9738

    2020

  54. [62]

    Van der Maaten, G

    L. Van der Maaten, G. Hinton, Visualizing data using t-sne., JMLR 9 (2008)

    2008

  55. [63]

    Martinetz, K

    T. Martinetz, K. Schulten, et al., A” neural-gas” network learns topolo- gies (1991). 38

    1991

Pith tools

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