Pith. sign in

REVIEW 4 major objections 7 minor 1 cited by

CALA: A Class-Aware Logit Adapter for Few-Shot Class-Incremental Learning

T0 review · 4 major / 7 minor · reviewed 2026-08-11 · deepseek-v4-flash

Pith's one-line read The paper claims that a class-aware logit adapter, trained on pseudo-incremental tasks, generates per-class logit balancing factors that correct confused novel classes, and reports state-of-the-art FSCIL accuracy on three benchmarks.

desk verdict Plausible plug-in idea for FSCIL, but the headline SOTA claim is confounded by an unmatched backbone and the theory section skips the actual loss. read the letter →

arxiv 2412.12654 v1 pith:LTSGT37X submitted 2024-12-17 cs.CV

classification cs.CV
keywords few-shotclass-incrementallearninglogitadjustmentclass-awareadapterprototypesimilaritymixuppseudo-incrementalplug-and-playmodulenovelclassconfusion
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

This paper tries to establish that the residual error in few-shot class-incremental learning (FSCIL) is a class-sensitive bias at the logit level, and that it can be corrected by a learned class-aware logit adapter. The typical incremental-frozen framework trains a backbone on abundant base classes and then freezes it, leaving novel classes systematically confused with the base classes they resemble. CALA adds a small adapter that, for each novel class, maps the cosine similarity between that class's prototype and all base-class prototypes to a scalar logit bonus; the bonus is added to the novel-class logit before the softmax. The adapter is trained offline in pseudo-incremental sessions whose fake novel classes are made by mixing base-class images, then frozen and applied to real incremental sessions. If the transfer from fake to real classes holds, CALA is a plug-and-play classifier-level correction for any frozen-backbone FSCIL method, and the paper reports consistent improvements on mini-ImageNet, CIFAR-100, and CUB-200.

What carries the argument

The central object is the class-aware logit adapter, a two-hidden-layer MLP $g_\phi$ that maps a normalized prototype-similarity vector to a single scalar. The similarity vector for a novel class $c$ is $S_c = [\mathrm{sim}(p_c,p_0^1),\dots,\mathrm{sim}(p_c,p_0^B)]$, where $p_c$ is the prototype of the novel class, $p_0^b$ are the frozen base-class prototypes in the classifier $W_0$, and $\mathrm{sim}$ is cosine similarity. The scalar output $\beta_c = g_\phi(\sigma(S_c))$ is added element-wise to the novel-class logits as $\hat{z} = z + [0,\gamma \beta]$, so that classes resembling base classes receive larger corrections. The machinery also includes the pseudo-incremental training loop: mixup fabricates fake novel classes from disjoint base-class pairs, their prototypes are computed with the frozen backbone, the MLP is trained to predict logit bonuses, and the loss $L_{\mathrm{CALA}} = -\sum_j \tilde{Y}_j \log \tilde{p}_j + \sum_c (\beta_c - \mu \|S_c\|_2)$ ties the bonus to the magnitude of the similarity vector. This coupling between similarity and bonus is what carries the transfer from pseudo-trained to real incremental sessions.

What would settle it

One concrete falsifier: build a FSCIL split whose novel classes are deliberately dissimilar from every base class (low cosine prototype similarity), and compare CALA's $\beta$ values against a class-agnostic constant. If the adapter still issues large bonuses to these dissimilar classes, or the accuracy gain vanishes, the learned similarity-to-bonus mapping does not generalize from mixup-fake to real novel classes.

Watch

Extended reading notes

Core claim

At the level of the paper's own argument, CALA's discovery is that the confusion between novel and base classes in FSCIL is not uniform across novel classes, so a single class-agnostic logit adjustment factor $\alpha$ is insufficient. The paper proposes that the required correction for a novel class $c$ should grow with its similarity to base classes, and implements this as a class-aware adapter $g_\phi$ that takes the similarity vector $S_c = [\cos(p_c,p_0^1),\dots,\cos(p_c,p_0^B)]$ and outputs a scalar $\beta_c$. The adjusted logit for the novel part is $\hat{z} = z + [0,\gamma \beta]$, with $\gamma$ controlling the overall strength. The adapter is trained by constructing fake novel classes via mixup of base-class instances, computing their prototypes, and optimizing a logit-adjustment loss whose simplified form $L_{\mathrm{CALA}} = \log(1+\sum_{k\neq j} e^{z_k-z_j})$ shows Fisher consistency: raising the target novel-class logit lowers the loss. In the real FSCIL stage the adapter is frozen and applied to genuine novel-class prototypes. On mini-ImageNet the method reports last-session accuracy 57.75% against a previous best of 54.04%, with gains also reported on CIFAR-100 and CUB-200.

Load-bearing premise

The load-bearing premise is that fake novel classes created by mixing base-class images trigger the same kind of classification mistakes as real novel classes, so the adapter's learned rule for how much to boost a class generalizes to classes it never saw.

Editorial extensions

If this is right

  • On mini-ImageNet, CALA raises last-session accuracy to 57.75%, up from the 54.04% of the strongest quoted baseline, and improves average accuracy over all sessions.
  • As a plug-and-play module, CALA improves three existing incremental-frozen methods (CEC, BiDist, SAVC) on mini-ImageNet, with gains concentrating in later sessions; the last-session harmonic mean of base and novel accuracy rises by 9.60, 9.19, and 3.97 points, respectively.
  • Because the correction is applied at the classifier level, the feature space and backbone are untouched, so CALA can be added to any prototype-based incremental-frozen FSCIL method without retraining the backbone.
  • The learned $\beta_c$ grows with novel-to-base similarity, matching the paper's motivation that more similar novel classes need stronger correction; the similarity analysis across datasets and within mini-ImageNet classes supports this monotone relationship.
  • The class-aware logit-adjustment loss is Fisher consistent: for a single novel class the loss simplifies to $\log(1+\sum_{k\neq j} e^{z_k-z_j})$, so raising the correct novel-class logit always decreases the loss.

Reading between the lines

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

  • Editorial inference: if the mixup-to-real transfer is robust, the same prototype-similarity-to-bonus mapping could be applied to other prototype-based continual learning settings with frozen backbones, such as online or task-free continual learning, where class confusion is likewise driven by feature-space overlap.
  • Editorial inference: the adapter's $\beta$ could double as an uncertainty signal; classes whose $\beta$ is high are those most entangled with base classes, so abstention or a separate rejection rule on those classes might further improve novel-class precision.
  • Editorial inference: the need to tune $\gamma$ suggests a testable extension in which the adapter also outputs $\gamma$, or $\gamma$ is set per session from the empirical similarity distribution, removing the global hyperparameter search.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 7 minor

Summary. The paper proposes CALA (Class-Aware Logit Adapter) for few-shot class-incremental learning. After training a frozen backbone on base classes, the method runs a pseudo-incremental stage in which fake novel classes are generated by mixup of base-class images. A two-layer MLP is trained to map the cosine-similarity vector between each fake novel prototype and the base prototypes to a scalar logit bonus beta_c; at real incremental sessions the frozen adapter computes beta for each novel class and adds gamma*beta to the novel-class logits before softmax. The manuscript claims that this class-aware logit adjustment reduces novel-class confusion, yields state-of-the-art accuracy on mini-ImageNet, CIFAR-100, and CUB-200, and serves as a plug-and-play module for existing incremental-frozen methods.

Significance. If the transfer from mixup-generated pseudo-novel classes to real novel classes holds, CALA would be a simple and useful classifier-level module that could be added to many FSCIL methods without retraining the backbone. The paper includes several useful diagnostic elements: an FPR analysis, confusion-matrix visualizations, t-SNE plots, and plug-and-play experiments on three baselines. However, the evidence as presented is not sufficient to support the central state-of-the-art claim. The main Table 2 comparison is confounded by an unmatched base-session accuracy gap, the mini-ImageNet SOTA margin largely disappears when CALA is added to a reproduced baseline in Table 4, the theoretical Fisher-consistency argument in Section 3.4 is derived from a loss that omits the actual adapter terms, and the similarity/regularizer analysis in Section 4.5 is partly circular because Eq. 9 directly enforces the reported relationship. The core idea remains plausible, but the paper currently overclaims what it demonstrates.

major comments (4)
  1. [§4.3, Table 2] The state-of-the-art comparison is not a controlled comparison of methods. CALA and LA-fscil both begin at 83.98% base-session accuracy, whereas the compared methods in Table 2 start at 72.17% to 80.47%. The claimed final-session margin (57.75% versus SAVC's 54.04%) is therefore largely inherited from the stronger backbone/base-training protocol rather than from the class-aware adapter. This is confirmed by Table 4, where adding CALA to the reproduced SAVC backbone improves the last session by only +1.02 points, to about 55.06%, which is below both LA-fscil (56.65%) and CALA's own headline number. To support the SOTA claim, the authors must compare against baselines trained under the same protocol or report gains relative to matched base-session accuracy.
  2. [§3.4, Eq. (9)–(11)] The Fisher-consistency analysis is not about CALA as defined. Equation (11) is obtained by dropping the logit-adjustment terms and the regularization term from Eq. (9), reducing the loss to log(1 + sum_{k neq j} e^{z_k - z_j}) for a single class. Moreover, z_j is produced by the frozen backbone and is not optimized by L_CALA; the trainable parameters are the adapter weights phi. The argument that increasing z_j decreases the loss therefore does not establish any property of the adapter or of the class-aware balancing factor. Either derive the consistency statement for the actual L_CALA including the beta terms and regularizer, or remove this theory subsection as unsupported.
  3. [§3.3, Eq. (1), and §3.4–3.5] The load-bearing transfer assumption is asserted but not validated. The adapter is trained on fake novel classes constructed by linearly mixing base-class images with lambda in [0.4, 0.6], and the same trained mapping is then applied to real incremental sessions. Because mixup features are convex combinations of base features, their cosine similarities to base prototypes may be systematically different from those of genuine novel classes, which would miscalibrate beta. The paper should provide direct evidence that the pseudo-task similarity geometry matches the real-task geometry, for example by comparing the distribution of novel-to-base similarity scores between pseudo-novel and real-novel classes, or by evaluating the adapter on held-out tasks before claiming generalization.
  4. [§4.5, Fig. 7, and §3.4, Eq. (9)] The empirical claim that larger novel-to-base similarity leads to larger beta_c is to a large degree enforced by the loss rather than discovered by the adapter. Equation (9) includes the regularization term sum_c (beta_c - mu ||S_c||_2), which explicitly drives beta_c toward a scaled version of the similarity norm. Presenting Fig. 7 as evidence that 'the greater the similarity, the larger the adjustment' is therefore circular unless the same relationship is shown when this regularizer is absent or its strength is varied. Please report an ablation with mu = 0 or otherwise separate the learned component of beta from the component imposed by the regularizer.
minor comments (7)
  1. [§3.3, Eq. (5)] The notation [0, alpha_tilde] is not defined; please state explicitly that 0 is a B-dimensional zero vector and that the bracket denotes concatenation, and clarify the distinction between the scalar alpha_c and the vector alpha_tilde.
  2. [§3.4, Eq. (6) and Eq. (9)] Equation (6) defines S_tilde_c as a vector of B cosine similarities, while Eq. (9) uses ||S_tilde_c||_2; please clarify the norm notation. Also, since the input to the MLP is softmax-normalized in Eq. (7) but the regularizer uses the L2 norm of the unnormalized vector, the relationship between the two quantities should be stated explicitly.
  3. [§4.2 and Algorithm 1] Several hyperparameters of the pseudo-training stage are missing: the number of pseudo-incremental sessions T_tilde, the number of pseudo-classes per session, the number of mixup samples per class, and the termination criterion for the 'while not done' loop in Algorithm 1. The 20 epochs mentioned in the text should be tied to the loop structure.
  4. [§4.3, Fig. 3] The CIFAR-100 and CUB-200 results are presented only as performance curves without numerical per-session values or standard deviations. Please provide tables or numeric listings so that the claimed consistent improvement can be checked.
  5. [§4.5, Table 4] The 'Improvement with CALA' rows report only deltas; the actual accuracies of the CALA-augmented baselines are not shown. This makes it impossible to see whether the final accuracies are close to the baselines, and the very large harmonic-mean gains should be accompanied by the separate base and novel accuracies used to compute them.
  6. [Figure 3 caption area] The text following the Figure 3 caption contains a long run of '/uni000000...' escape-like tokens. If this text is present in the submitted manuscript it must be removed; if it is an extraction artifact, the figure and caption should be re-checked in the source file.
  7. [§4.4, Fig. 4] The statement that the red lines 'better fit the black dashed line' after CALA is a visual claim. Please add a quantitative measure of logit imbalance, such as the average base-versus-novel logit gap, rather than relying on inspection of the plot.

Circularity Check

3 steps flagged · score 4.0 of 10

The class-aware adapter's own supporting evidence (similarity-β trend and FPR reduction) is circular by construction, while the central benchmark evaluation is not; the SOTA comparison is confounded rather than circular.

  1. self definitional [Sec. 3.4, Eq. (9); Sec. 4.5, 'Similarity analysis', Fig. 7]
    "LCALA = −Σ_j Y˜j log(p˜j) + Σ_c (β_c − µ||S˜c||2). (9) ... When novel classes are more similar with base classes, the mean value of β_c increases, as we assign higherβ_c to novel classes that are more similar to base classes for stronger adjustment."

    The second term of Eq. (9) is a regularizer that drives β_c toward μ·||S̃_c||_2, making β_c an affine function of the similarity norm during training. The adapter is trained by minimizing this objective, so the positive β-to-similarity relationship reported in Fig. 7 is a direct consequence of the loss, not an independent empirical discovery. The paper presents Fig. 7 as a 'Further study' that validates the motivation that more similar novel classes need larger rectification, but that relationship is already inscribed in the training objective by construction.

  2. self definitional [Table 1 and Sec. 4.3 'Comparisons with SOTAs', with Eq. (8)]
    "As shown in Table 1, our methods have lower FPR than other baselines in each session, namely, they are less likely to confuse novel instances into base classes. ... In conclusion, CALA fully aligns with its motivation and is relatively more effective."

    Eq. (8) adds a positive vector γβ to the novel-class logits while leaving base logits unchanged. FPR is the fraction of novel instances predicted into base classes, so raising all novel logits by a positive amount mechanically lowers FPR, regardless of whether overall accuracy improves. The FPR comparison therefore does not independently validate CALA; it is the same arithmetic operation the method performs. The paper's claim that this 'fully aligns with its motivation' is a validation by construction rather than evidence of effectiveness.

1 more flagged steps
  1. other [Sec. 3.4, 'Theory analysis', Eq. (11)]
    "Assuming a novel class j, our method aims to enhance the logit of novel classes to make the decision boundary more favorable to novel classes, which corresponds with the increase of zj and the decrease of the loss function, proving the Fisher consistency of our class-aware logit adapter loss."

    The simplification to Eq. (11) is just the cross-entropy loss for class j with adjusted logits; any such loss decreases when the correct-class logit z_j increases. Fisher consistency is a statement about the expected-loss minimizer recovering the Bayes-optimal decision rule under the true label distribution, which this monotonicity does not establish. The 'proof' therefore renames a trivial property of the loss as a statistical consistency guarantee, offering no independent mathematical support for the calibration.

full rationale

The core adaptation mechanism is not circular in the narrowest sense: the β values used for real novel classes are produced by a frozen adapter trained on mixup-generated pseudo-novel classes, rather than fitted to the labels of the incremental test sessions. However, two pieces of evidence offered for the method's core intuition are circular by construction. Eq. (9) contains an explicit regularization term (β_c − μ||S̃_c||_2), so the positive similarity-β trend in Fig. 7 is imposed by the loss rather than discovered. Likewise, the FPR table 'validates' CALA with a metric that Eq. (8) directly manipulates: adding positive β to novel logits mechanically reduces the chance that novel instances are classified into base classes. The Fisher-consistency proof in Sec. 3.4 is also tautological, reducing consistency to the trivial fact that increasing the correct logit decreases a cross-entropy-style loss. The headline SOTA comparison is not circular but is empirically confounded: CALA and LA-fscil inherit a stronger backbone (83.98% session-0 accuracy) than most reproduced baselines, and LA-fscil alone already exceeds the prior SOTA, so the claim that experiments 'consistently validate the effectiveness' of the class-aware adapter is not a controlled demonstration. Because the benchmark accuracy itself is evaluated on held-out novel classes and does not reduce to the training objective, the appropriate circularity score is moderate rather than severe.

Assumptions & free parameters 4 free parameters · 3 assumptions · 0 invented entities

No new physical or ontological entities. The adapter is a two-layer MLP and beta is a scalar multiplier; both are standard computational constructs. The paper's novelty lies in how they are used, not in postulating new entities.

free parameters (4)
  • gamma (scaling factor) = 10 (CIFAR-100; other datasets unspecified)
    Scales the class-aware bonus in Eq. 8. Ablation in Table 3 shows sensitivity: gamma=1 gives no gain, gamma=100 collapses accuracy to 26.53%.
  • mu (regularization coefficient) = not reported
    Regularization coefficient in Eq. 9 that ties beta to the similarity norm. Its value is never given, so the training objective is not reproducible.
  • MLP hidden units = 64
    Adapter architecture choice for the two hidden layers; not justified or ablated.
  • pseudo-session count T_tilde = not specified
    Algorithm 1 loops over T_tilde pseudo-sessions, but the paper never states the value used in the experiments.
assumptions (3)
  • domain assumption The backbone is frozen after base training and is never updated on novel classes.
    Section 3.2 states the backbone is kept frozen; this is the standard FSCIL incremental-frozen paradigm the paper builds on.
  • domain assumption Cosine similarity between class prototypes captures the degree of confusion between a novel class and base classes.
    Eq. 6 and the adapter input rely on this; no proof is given, and the relationship is only verified post hoc on the same data used to train the adapter.
  • ad hoc to paper Fake novel classes created by mixup of base classes have confusion properties similar to real novel classes.
    Eq. 1 and the pseudo-training paradigm in Section 3.4; this is the key transfer assumption that makes the adapter usable in real sessions.

how reviews work

0 comments
Cite this review

Pith. "Pith review of CALA: A Class-Aware Logit Adapter for Few-Shot Class-Incremental Learning." pith.science (2026). https://pith.science/paper/LTSGT37X

@misc{pith2026241212654,
  author       = {Pith},
  title        = {Pith review of: CALA: A Class-Aware Logit Adapter for Few-Shot Class-Incremental Learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/LTSGT37X}},
  note         = {Machine review of arXiv:2412.12654}
}
read the original abstract

Few-Shot Class-Incremental Learning (FSCIL) defines a practical but challenging task where models are required to continuously learn novel concepts with only a few training samples. Due to data scarcity, existing FSCIL methods resort to training a backbone with abundant base data and then keeping it frozen afterward. However, the above operation often causes the backbone to overfit to base classes while overlooking the novel ones, leading to severe confusion between them. To address this issue, we propose Class-Aware Logit Adapter (CALA). Our method involves a lightweight adapter that learns to rectify biased predictions through a pseudo-incremental learning paradigm. In the real FSCIL process, we use the learned adapter to dynamically generate robust balancing factors. These factors can adjust confused novel instances back to their true label space based on their similarity to base classes. Specifically, when confusion is more likely to occur in novel instances that closely resemble base classes, greater rectification is required. Notably, CALA operates on the classifier level, preserving the original feature space, thus it can be flexibly plugged into most of the existing FSCIL works for improved performance. Experiments on three benchmark datasets consistently validate the effectiveness and flexibility of CALA. Codes will be available upon acceptance.

Figures

Figures reproduced from arXiv: 2412.12654 by the authors.

Figure 1
Figure 1. Comparisons of (a) previous works with the incremental [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. An overview of our method CALA. (a) In the base session, we use sufficient base data to pre-train a generalizable backbone that [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Comparison with the state-of-the-art works on the other [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Ablation study on the logit distribution before and after [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 6
Figure 6. Figure 6: t-SNE [30] visualization of feature space without and with CALA. Dots and triangles represent instances from base and novel classes separately. Different colors represent differ￾ent classes. Stars indicate the centroids of classes. Geometric backgrounds with correspond…
Figure 7
Figure 7. Figure 7: Further study on the relationship between similarity and [PITH_FULL_IMAGE:figures/full_fig_p009_7.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Ambiguity-Guided Learnable Distribution Calibration for Semi-Supervised Few-Shot Class-Incremental Learning

    cs.CV 2025-07 reject novelty 5.0 of 10

    Under a new unlabeled-set composition that mixes base and novel classes, the ALDC method reports average accuracy of 69.77% on mini-ImageNet, 10.9 to 11.5 points above prior Semi-FSCIL baselines.

Reference graph

Works this paper leans on

48 extracted references · 29 canonical work pages · cited by 1 Pith paper

  1. [1]

    Learning to learn task-adaptive hyperparameters for few-shot learning

    Sungyong Baik, Myungsub Choi, Janghoon Choi, Heewon Kim, and Kyoung Mu Lee. Learning to learn task-adaptive hyperparameters for few-shot learning. IEEE Transactions on Pattern Analysis and Machine Intelligence , 46(3):1441– 1454, 2023. 2

  2. [2]

    Dark experience for gen- eral continual learning: a strong, simple baseline

    Pietro Buzzega, Matteo Boschini, Angelo Porrello, Davide Abati, and Simone Calderara. Dark experience for gen- eral continual learning: a strong, simple baseline. Advances in neural information processing systems, 33:15920–15930,

  3. [3]

    Learning imbalanced datasets with label- distribution-aware margin loss

    Kaidi Cao, Colin Wei, Adrien Gaidon, Nikos Arechiga, and Tengyu Ma. Learning imbalanced datasets with label- distribution-aware margin loss. Advances in neural informa- tion processing systems, 32, 2019. 2, 5

  4. [4]

    Metafscil: A meta-learning approach for few-shot class incremental learning

    Zhixiang Chi, Li Gu, Huan Liu, Yang Wang, Yuanhao Yu, and Jin Tang. Metafscil: A meta-learning approach for few-shot class incremental learning. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 14166–14175, 2022. 2

  5. [5]

    Imagenet: A large-scale hierarchical image database

    Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. Imagenet: A large-scale hierarchical image database. In 2009 IEEE conference on computer vision and pattern recognition, pages 248–255. Ieee, 2009. 1

  6. [6]

    Arcface: Additive angular margin loss for deep face recognition

    Jiankang Deng, Jia Guo, Niannan Xue, and Stefanos Zafeiriou. Arcface: Additive angular margin loss for deep face recognition. In Proceedings of the IEEE/CVF con- ference on computer vision and pattern recognition , pages 4690–4699, 2019. 2

  7. [7]

    Model- agnostic meta-learning for fast adaptation of deep networks

    Chelsea Finn, Pieter Abbeel, and Sergey Levine. Model- agnostic meta-learning for fast adaptation of deep networks. In International conference on machine learning , pages 1126–1135. PMLR, 2017. 2

  8. [8]

    An empirical investigation of catas- trophic forgetting in gradient-based neural networks

    Ian J Goodfellow, Mehdi Mirza, Da Xiao, Aaron Courville, and Yoshua Bengio. An empirical investigation of catas- trophic forgetting in gradient-based neural networks. arXiv preprint arXiv:1312.6211, 2013. 1

Show all 48 references
  1. [9]

    Mamba: Linear-time sequence modeling with selective state spaces

    Albert Gu and Tri Dao. Mamba: Linear-time sequence modeling with selective state spaces. arXiv preprint arXiv:2312.00752, 2023. 1

  2. [10]

    Adaptive distribution calibration for few- shot learning with hierarchical optimal transport

    Dandan Guo, Long Tian, He Zhao, Mingyuan Zhou, and Hongyuan Zha. Adaptive distribution calibration for few- shot learning with hierarchical optimal transport. Advances in neural information processing systems , 35:6996–7010,

  3. [11]

    Deep residual learning for image recognition

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

  4. [12]

    Mask r-cnn

    Kaiming He, Georgia Gkioxari, Piotr Doll ´ar, and Ross Gir- shick. Mask r-cnn. In Proceedings of the IEEE international conference on computer vision, pages 2961–2969, 2017. 1

  5. [13]

    A theoretical study on solving continual learn- ing

    Gyuhak Kim, Changnan Xiao, Tatsuya Konishi, Zixuan Ke, and Bing Liu. A theoretical study on solving continual learn- ing. Advances in neural information processing systems, 35: 5065–5079, 2022. 1

  6. [14]

    Learning multiple layers of features from tiny images

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

  7. [15]

    Unsupervised model personalization while preserving privacy and scalabil- ity: An open problem

    Matthias De Lange, Xu Jia, Sarah Parisot, Ales Leonardis, Gregory Slabaugh, and Tinne Tuytelaars. Unsupervised model personalization while preserving privacy and scalabil- ity: An open problem. In Proceedings of the IEEE/CVF con- ference on computer vision and pattern recognit...

  8. [16]

    Adversarial feature hallucination networks for few-shot learning

    Kai Li, Yulun Zhang, Kunpeng Li, and Yun Fu. Adversarial feature hallucination networks for few-shot learning. In Pro- ceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 13470–13479, 2020. 2

  9. [17]

    Adjusting logit in gaussian form for long-tailed visual recognition

    Mengke Li, Yiu-ming Cheung, Yang Lu, Zhikai Hu, We- ichao Lan, and Hui Huang. Adjusting logit in gaussian form for long-tailed visual recognition. IEEE Transactions on Ar- tificial Intelligence, 2024. 2

  10. [18]

    Feature space transfer for data augmen- tation

    Bo Liu, Xudong Wang, Mandar Dixit, Roland Kwitt, and Nuno Vasconcelos. Feature space transfer for data augmen- tation. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 9090–9098, 2018. 2

  11. [19]

    Swin transformer: Hierarchical vision transformer using shifted windows

    Ze Liu, Yutong Lin, Yue Cao, Han Hu, Yixuan Wei, Zheng Zhang, Stephen Lin, and Baining Guo. Swin transformer: Hierarchical vision transformer using shifted windows. In Proceedings of the IEEE/CVF international conference on computer vision, pages 10012–10022, 2021. 1

  12. [20]

    Generative ad- versarial residual pairwise networks for one shot learning

    Akshay Mehrotra and Ambedkar Dukkipati. Generative ad- versarial residual pairwise networks for one shot learning. arXiv preprint arXiv:1703.08033, 2017. 2

  13. [21]

    Long-tail learning via logit adjustment

    Aditya Krishna Menon, Sadeep Jayasumana, Ankit Singh Rawat, Himanshu Jain, Andreas Veit, and Sanjiv Kumar. Long-tail learning via logit adjustment. arXiv preprint arXiv:2007.07314, 2020. 2

  14. [22]

    Few-shot class-incremental learning from an open- set perspective

    Can Peng, Kun Zhao, Tianren Wang, Meng Li, and Brian C Lovell. Few-shot class-incremental learning from an open- set perspective. In European Conference on Computer Vi- sion, pages 382–397. Springer, 2022. 7

  15. [23]

    icarl: Incremental classifier and representation learning

    Sylvestre-Alvise Rebuffi, Alexander Kolesnikov, Georg Sperl, and Christoph H Lampert. icarl: Incremental classifier and representation learning. In Proceedings of the IEEE con- ference on Computer Vision and Pattern Recognition, pages 2001–2010, 2017. 7 9

  16. [24]

    Imagenet large scale visual recognition challenge

    Olga Russakovsky, Jia Deng, Hao Su, Jonathan Krause, San- jeev Satheesh, Sean Ma, Zhiheng Huang, Andrej Karpathy, Aditya Khosla, Michael Bernstein, et al. Imagenet large scale visual recognition challenge. International journal of computer vision, 115:211–252, 2015. 5

  17. [25]

    Prototypical networks for few-shot learning

    Jake Snell, Kevin Swersky, and Richard Zemel. Prototypical networks for few-shot learning. Advances in neural informa- tion processing systems, 30, 2017. 2

  18. [26]

    Learning with fantasy: Semantic-aware virtual contrastive constraint for few-shot class-incremental learning

    Zeyin Song, Yifan Zhao, Yujun Shi, Peixi Peng, Li Yuan, and Yonghong Tian. Learning with fantasy: Semantic-aware virtual contrastive constraint for few-shot class-incremental learning. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages...

  19. [27]

    Pseudo re- hearsal using non photo-realistic images

    Bhasker Sri Harsha Suri and Kalidas Yeturu. Pseudo re- hearsal using non photo-realistic images. In 2020 25th Inter- national Conference on Pattern Recognition (ICPR) , pages 4797–4804. IEEE, 2021. 1

  20. [28]

    Few-shot class- incremental learning

    Xiaoyu Tao, Xiaopeng Hong, Xinyuan Chang, Songlin Dong, Xing Wei, and Yihong Gong. Few-shot class- incremental learning. In Proceedings of the IEEE/CVF con- ference on computer vision and pattern recognition , pages 12183–12192, 2020. 1, 2, 5, 7

  21. [29]

    Local and global logit adjustments for long-tailed learning

    Yingfan Tao, Jingna Sun, Hao Yang, Li Chen, Xu Wang, Wenming Yang, Daniel Du, and Min Zheng. Local and global logit adjustments for long-tailed learning. InProceed- ings of the IEEE/CVF International Conference on Com- puter Vision, pages 11783–11792, 2023. 2

  22. [30]

    Visualizing data using t-sne

    Laurens Van der Maaten and Geoffrey Hinton. Visualizing data using t-sne. Journal of machine learning research , 9 (11), 2008. 8

  23. [31]

    Attention is all you need

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszko- reit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. Attention is all you need. Advances in neural information processing systems, 30, 2017. 1

  24. [32]

    The caltech-ucsd birds-200-2011 dataset

    Catherine Wah, Steve Branson, Peter Welinder, Pietro Per- ona, and Serge Belongie. The caltech-ucsd birds-200-2011 dataset. 2011. 5

  25. [33]

    Cosface: Large margin cosine loss for deep face recognition

    Hao Wang, Yitong Wang, Zheng Zhou, Xing Ji, Dihong Gong, Jingchao Zhou, Zhifeng Li, and Wei Liu. Cosface: Large margin cosine loss for deep face recognition. In Pro- ceedings of the IEEE conference on computer vision and pat- tern recognition, pages 5265–5274, 2018. 2

  26. [34]

    Few-shot class-incremental learning via training-free prototype calibration

    Qi-Wei Wang, Da-Wei Zhou, Yi-Kai Zhang, De-Chuan Zhan, and Han-Jia Ye. Few-shot class-incremental learning via training-free prototype calibration. Advances in Neural Information Processing Systems, 36, 2024. 2, 5, 6, 7

  27. [35]

    Margin calibration for long-tailed visual recognition

    Yidong Wang, Bowen Zhang, Wenxin Hou, Zhen Wu, Jin- dong Wang, and Takahiro Shinozaki. Margin calibration for long-tailed visual recognition. In Asian Conference on Ma- chine Learning, pages 1101–1116. PMLR, 2023. 2

  28. [36]

    A unified generaliza- tion analysis of re-weighting and logit-adjustment for imbal- anced learning

    Zitai Wang, Qianqian Xu, Zhiyong Yang, Yuan He, Xi- aochun Cao, and Qingming Huang. A unified generaliza- tion analysis of re-weighting and logit-adjustment for imbal- anced learning. Advances in Neural Information Processing Systems, 36, 2024. 2

  29. [37]

    Learning imbalanced data with vision transformers

    Zhengzhuo Xu, Ruikang Liu, Shuo Yang, Zenghao Chai, and Chun Yuan. Learning imbalanced data with vision transformers. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages 15793– 15803, 2023. 2

  30. [38]

    Scrollnet: Dy- namicweight importance for continual learning

    Fei Yang, Kai Wang, and Joost van de Weijer. Scrollnet: Dy- namicweight importance for continual learning. In Proceed- ings of the IEEE/CVF International Conference on Com- puter Vision, pages 3345–3355, 2023. 1

  31. [39]

    Free lunch for few- shot learning: Distribution calibration

    Shuo Yang, Lu Liu, and Min Xu. Free lunch for few- shot learning: Distribution calibration. arXiv preprint arXiv:2101.06395, 2021. 2

  32. [40]

    Neural collapse inspired feature- classifier alignment for few-shot class incremental learning

    Yibo Yang, Haobo Yuan, Xiangtai Li, Zhouchen Lin, Philip Torr, and Dacheng Tao. Neural collapse inspired feature- classifier alignment for few-shot class incremental learning. arXiv preprint arXiv:2302.03004, 2023. 2

  33. [41]

    Few-shot incremental learning with contin- ually evolved classifiers

    Chi Zhang, Nan Song, Guosheng Lin, Yun Zheng, Pan Pan, and Yinghui Xu. Few-shot incremental learning with contin- ually evolved classifiers. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages 12455–12464, 2021. 2, 3, 6, 7, 8

  34. [42]

    mixup: Beyond empirical risk minimiza- tion

    Hongyi Zhang, MitMoustapha Cisse, YannN Dauphin, and David Lopez-Paz. mixup: Beyond empirical risk minimiza- tion. 2017. 3

  35. [43]

    Class-incremental learning via deep model consolidation

    Junting Zhang, Jie Zhang, Shalini Ghosh, Dawei Li, Serafet- tin Tasci, Larry Heck, Heming Zhang, and C-C Jay Kuo. Class-incremental learning via deep model consolidation. In Proceedings of the IEEE/CVF winter conference on applica- tions of computer vision, pages 1131–1140, 2020. 1

  36. [44]

    Few-shot class- incremental learning via class-aware bilateral distillation

    Linglan Zhao, Jing Lu, Yunlu Xu, Zhanzhan Cheng, Dashan Guo, Yi Niu, and Xiangzhong Fang. Few-shot class- incremental learning via class-aware bilateral distillation. In Proceedings of the IEEE/CVF conference on computer vi- sion and pattern recognition, pages 11838–11847, 202...

  37. [45]

    Forward compatible few-shot class-incremental learning

    Da-Wei Zhou, Fu-Yun Wang, Han-Jia Ye, Liang Ma, Shil- iang Pu, and De-Chuan Zhan. Forward compatible few-shot class-incremental learning. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages 9046–9056, 2022. 1, 6, 7

  38. [46]

    Few-shot class-incremental learn- ing by sampling multi-phase tasks

    Da-Wei Zhou, Han-Jia Ye, Liang Ma, Di Xie, Shiliang Pu, and De-Chuan Zhan. Few-shot class-incremental learn- ing by sampling multi-phase tasks. IEEE Transactions on Pattern Analysis and Machine Intelligence , 45(11):12816– 12831, 2022. 7

  39. [47]

    Self-promoted prototype refinement for few-shot class- incremental learning

    Kai Zhu, Yang Cao, Wei Zhai, Jie Cheng, and Zheng-Jun Zha. Self-promoted prototype refinement for few-shot class- incremental learning. In Proceedings of the IEEE/CVF con- ference on computer vision and pattern recognition , pages 6801–6810, 2021. 1, 2

  40. [48]

    Gkeal: Gaussian kernel embedded analytic learning for few-shot class incremental task

    Huiping Zhuang, Zhenyu Weng, Run He, Zhiping Lin, and Ziqian Zeng. Gkeal: Gaussian kernel embedded analytic learning for few-shot class incremental task. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 7746–7755, 2023. 7 10

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.