Pith. sign in

REVIEW 3 major objections 5 minor 88 references

Class Balance Matters to Active Class-Incremental Learning

T0 review · 3 major / 5 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read Class-balanced selection outperforms random and active learning baselines in the new Active Class-Incremental Learning protocol.

desk verdict A genuinely new task (ACIL) and a simple class-balance-driven selector, but the main comparison is confounded by mismatched training schedules: CBS gets 50 epochs once, baselines get up to 130+ epochs in multi-round loops. read the letter →

arxiv 2412.06642 v1 pith:XRJIPFGJ submitted 2024-12-09 cs.CV

Zitong Huang , Ze Chen , Yuanze Li , Bowen Dong , Erjin Zhou , Yong Liu , Rick Siow Mong Goh , Chun-Mei Feng
show 1 more author
Wangmeng Zuo
This is my paper · ORCID
classification cs.CV
keywords ActiveClass-IncrementalLearningclass-balancedselectionfew-shotprompttuningk-meansclusteringGaussiandistributionmatchingclass-imbalancedratio
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 proposes Active Class-Incremental Learning (ACIL), in which each incremental session begins with a large pool of unlabeled images and a small labeling budget, and the goal is to choose which images to label so that the incremental learner performs as well as possible. The authors report that standard active learning heuristics—uncertainty, diversity, and hybrid scores—select class-imbalanced batches under this protocol and can do worse than random sampling, because an imbalanced training set hurts the incremental learner. They propose Class-Balanced Selection (CBS), which clusters the unlabeled features into as many groups as there are classes and then greedily picks samples from each cluster so that the selected samples' Gaussian distribution matches the cluster's distribution. Across five datasets and three prompt-tuning incremental methods, CBS beats random selection, the active learning baselines, and forced balanced random selection at the same labeling budget, with the largest gains at small budgets.

What carries the argument

The core object is the Class-Balanced Selection (CBS) routine. At each session it extracts normalized features with the pretrained encoder, runs k-means with k equal to the session's class count, allocates selections to each cluster in proportion to cluster size, and then greedily adds samples that minimize the KL divergence between the Gaussian fitted to the selected samples and the Gaussian fitted to the entire cluster. This distribution-matching step buys both class balance, through proportional cluster budgets, and representativeness, through per-cluster greedy matching. For LP-DiF, a second mechanism estimates per-class Gaussians from labeled plus pseudo-labeled unselected data, so replay pseudo-features better resist catastrophic forgetting.

What would settle it

Build an incremental session from two classes whose pretrained features overlap heavily, run CBS with the same budget, and measure the class-imbalanced ratio of the selected set and the session's accuracy. If the ratio is no better than random and accuracy drops to the random baseline, the cluster-alignment premise fails.

Watch

Extended reading notes

Core claim

The central claim is that class balance, not uncertainty or diversity per se, is what makes an annotated subset useful for incremental learning under the ACIL protocol. The paper shows empirically that existing active learning methods (Entropy, Margin, Coreset, BADGE, Typiclust, ProbCover, DropQuery) produce class-imbalanced selections that are sometimes worse than random selection, and that their accuracy trails random selection as a result. CBS fixes this by making the selected set's feature distribution mirror the whole unlabeled pool: k-means splits the pool into as many clusters as there are classes, each cluster contributes a number of samples proportional to its size, and within each cluster a greedy rule adds the sample that most reduces the KL divergence between the Gaussian of the selected samples and the Gaussian of the cluster. When plugged into L2P, DualPrompt, and LP-DiF, CBS raises average session accuracy over random and all baselines on five datasets; in LP-DiF, the paper further uses the unselected unlabeled data to estimate class Gaussians for pseudo-feature replay, which specifically improves old-class accuracy.

Load-bearing premise

The method assumes k-means clusters of pretrained features line up with the true classes in each session's unlabeled pool, so that sampling proportionally from each cluster yields a class-balanced labeled set.

Editorial extensions

If this is right

  • At a labeling budget of $B=100$, CBS raises mean average accuracy over five datasets to 81.42 for L2P, 83.01 for DualPrompt, and 82.03 for LP-DiF, versus 78.19, 79.77, and 80.45 for random selection.
  • The advantage over random and prior active learning grows at small budgets ($B=40$ or $60$), where class imbalance is most damaging and CBS's classes-discovery ratio stays highest.
  • CBS is model-agnostic for prompt-tuning incremental learners: it swaps into L2P, DualPrompt, and LP-DiF without changing their training procedures.
  • For LP-DiF, incorporating unselected unlabeled data into Gaussian replay improves old-class accuracy (from 66.21% to 67.66% on the last CUB-200 session) while leaving new-class accuracy unchanged.
  • CBS selects a batch in one pass, so it is cheaper than multi-round active learners: 42 seconds per session versus 149 for DropQuery on CUB-200.

Reading between the lines

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

  • Beyond the paper's non-overlapping-class assumption, the method's balance guarantee needs the clusters to track true classes; an open-world ACIL setting would require estimating the number of clusters or filtering old-class features before running CBS.
  • The greedy step is pure distribution matching on frozen features, so the same routine could be reused for any budgeted subset-selection task where representativeness matters, such as coreset construction, without retraining.
  • The paper's link between class-imbalanced ratio and accuracy suggests a cheap diagnostic: measure the imbalance ratio of any proposed selection before labeling, and if it is worse than random, expect the learner to underperform random.
  • The LP-DiF gain from pseudo-labeling the unused pool suggests active selection and semi-supervised distribution estimation are complementary; combining CBS with confidence-thresholded pseudo-labeling is a testable extension.
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

3 major / 5 minor

Summary. The paper introduces Active Class-Incremental Learning (ACIL), where each incremental session has a large unlabeled pool and a labeling budget. It proposes Class-Balanced Selection (CBS), which extracts features with a frozen pretrained encoder, clusters the unlabeled pool into |C_t| groups via k-means, and then greedily selects samples within each cluster so that the Gaussian distribution of the selected features matches the Gaussian distribution of the cluster. When used with LP-DiF, the method also exploits the unselected unlabeled data to improve Gaussian replay distributions. The method is evaluated on five datasets with three prompt-based CIL backbones (L2P, DualPrompt, LP-DiF) and compared against random selection and seven active learning baselines; the paper reports that CBS outperforms both random and the baselines.

Significance. If the empirical claims hold, ACIL is a useful problem formulation and CBS is a practical, model-agnostic selection strategy that could reduce annotation cost while preserving incremental learning accuracy. The strengths include the public code release, evaluation across five datasets and three CIL backbones, and an ablation (Table 2) that separates the clustering step from the greedy selection step. However, the central empirical claim is currently undermined by a protocol confound in the comparison with active learning baselines, and the paper's supporting analysis of class balance relies on an unvalidated cluster-alignment assumption and a circular KL-divergence demonstration. The core idea remains plausible, but the evidence as reported does not yet isolate the effect of the selection strategy.

major comments (3)
  1. [Sec. 4.1, Implementation Details; Table 1; Fig. 2] The comparison between CBS and the active learning baselines is confounded by training protocol. CBS selects all B samples at once and trains for 50 epochs, whereas the AL baselines are run in a multi-round loop: 20 samples selected, labeled, and trained per round, with 20 epochs in each of the first R-1 rounds and 50 epochs in the final round. For B=100, this gives the baselines 130 epochs versus CBS's 50 epochs. The reported Avg gains for CBS could therefore be caused by the single-shot training schedule rather than by the selection quality. This directly impacts the headline claim in the Abstract and Section 1 that CBS outperforms existing AL approaches. The authors should rerun the comparison with the training schedule equalized across both arms (e.g., train all selected sets for the same total number of epochs, or give CBS the same multi-round training as the baselines) and re-report Table 1 and Fig. 2 accordingly.
  2. [Alg. 2, line 4; Sec. 3.2; Sec. A of the appendix] The class-balance guarantee of CBS rests on the assumption that k-means clustering with k = |C_t| produces clusters that align with the true session classes. The paper provides no evidence of this alignment: the 'class-imbalanced ratio' analysis in the appendix is computed using ground-truth labels of the selected samples, not on the cluster composition, and no cluster-purity metric is reported. If the clusters do not correspond to classes, proportional sampling per cluster does not guarantee class balance, and the KL objective only matches the selected set to the cluster distribution, not to the class distribution. Please report cluster purity (e.g., adjusted Rand index against the session's class labels) or vary k and show the sensitivity; without this, the mechanism linking CBS to class balance is unsupported.
  3. [Appendix D, Eq. (4) and Fig. 7] The KL-divergence comparison in Fig. 7 is circular as a demonstration of representativeness, because CBS greedily minimizes exactly this KL divergence between the selected set's Gaussian and the full cluster's Gaussian (Eq. 2 in Sec. 3.2). Reporting that CBS achieves lower KL than random is therefore a direct consequence of the optimization objective, not independent evidence that the selected samples are more representative. If this figure is meant as a sanity check of the optimizer, it should be labeled as such; otherwise, please provide an external metric not aligned with the objective, such as nearest-neighbor classification accuracy on the selected set or coverage of the feature space.
minor comments (5)
  1. [Appendix A] The sentence 'CIFAR-100 consists of 100 general classes, each of which contains 50,000 training images' is incorrect as written (CIFAR-100 has 50,000 training images in total); it should say each class contains about 500 training images.
  2. [Appendix D, Table 4] The cross-reference to Table 4 appears as 'Tab. ??' in the text; please fix the reference.
  3. [Sec. 4.2, after Table 1] The phrase 'our CBS outperform Balance random for all CIL methods on most datasets' contains a typo ('outperform' should be 'outperforms') and the term 'Balance random' is not defined in the main text; the table uses 'Balanced random (FSCIL)'.
  4. [Sec. A of the appendix, 'class-imbalanced ratio' definition] The class-imbalanced ratio divides the count of the most-selected class by the count of the least-selected class; please state explicitly how ties and classes with zero selected samples are handled, since those cases affect the ratio significantly when the budget is small.
  5. [Sec. 4.3, Table 2] The ablation that replaces the greedy selection with Entropy, Coreset, and BADGE within each cluster is reported only on CUB-200 with LP-DiF at B=100; adding at least one more dataset/backbone would strengthen the claim that the greedy step, not the clustering, is responsible for the improvement.

Circularity Check

1 steps flagged · score 3.0 of 10

One supporting analysis is circular because it reports the selection objective itself as evidence of representativeness; the central test-accuracy claim is externally measured and remains non-circular.

  1. self definitional [Supplementary Material Sec. D, Eq. (4) / Fig. 7; compare Alg. 2 lines 10-13 and Eq. (2)]
    "we calculate the KL divergence between the Gaussian distribution of the selected samples for each class and the distribution of all samples of that class in the unlabeled pool, using the following formula: ... Clearly, the samples selected by our CBS have a lower KL divergence with the entire sample set of most classes compared to those selected by random selection. This demonstrates that our method indeed ensures that the selected samples are more representative of the overall distribution."

    Algorithm 2 line 11 selects each sample to minimize D_KL(N(mu_j, sigma^2_j) | P(S_j union {f})), i.e., the selected set is constructed to minimize exactly the KL divergence in Eq. (2). Appendix D then evaluates Eq. (4), which is the same formula, and presents CBS's lower value as a demonstration that the selected samples are more representative. The metric is the algorithm's own objective, so this is a self-definitional restatement, not an independent confirmation. The main test-accuracy results, however, are measured on held-out data and are not reduced by this step.

full rationale

The central performance claim, that CBS improves incremental test accuracy over random selection and active-learning baselines, rests on held-out Avg accuracy (Table 1, Fig. 2, and the budget curves), which is an external quantity not optimized by the selection rule. That comparison is self-contained and not circular. The one exhibited circular step is the supporting analysis in Appendix D: Eq. (4) is identical to Eq. (2), and Alg. 2 greedily minimizes that same KL divergence, so reporting that CBS achieves a lower KL value is a restatement of the selection objective rather than an independent demonstration of representativeness. This is a non-central, definitional circularity. The self-citation to LP-DiF [26] is used as a base incremental method and motivation for Gaussian replay, but it is not load-bearing: the same plug-and-play evidence is provided with external methods L2P and DualPrompt, and the CBS improvement over LP-DiF is again measured by test accuracy. Two experimental-design concerns are correctness risks rather than circularity: the asymmetric training schedules (CBS trains 50 epochs once; multi-round baselines train 20 epochs in the first R-1 rounds and 50 in the last, Appendix B) and the reliance on k-means clusters aligning with true classes. Neither reduces the result to its inputs, so they do not increase the circularity score beyond 3.

Assumptions & free parameters 1 free parameters · 3 assumptions · 0 invented entities

The method introduces no new physical entities. Its reliance is on three key assumptions: disjoint new-class pools, cluster-class alignment from k-means, and Gaussian-distributed features within clusters. The number of clusters is set to the class count, which is a hand-chosen design parameter. These assumptions are not fitted to data but are load-bearing for the class-balance guarantee.

free parameters (1)
  • k (number of clusters) = |C_t| (number of classes in session)
    The k-means algorithm is run with k equal to the known number of classes in the session's unlabeled pool. This is a design choice, not learned from data, but it assumes the cluster structure aligns with class structure.
assumptions (3)
  • domain assumption The unlabeled pool in each session contains only classes not seen before (class spaces are non-overlapping).
    Stated in Sec. 3 Problem Formulation: 'for different sessions, the class spaces are non-overlapping.' The authors acknowledge in the Limitations that this is hard to satisfy in real applications.
  • ad hoc to paper k-means with k = |C_t| produces clusters that correspond to the true classes.
    Algorithm 2 line 4. The method relies on each cluster being roughly a class so that proportional sampling from clusters yields class balance. No validation of cluster-class correspondence is provided.
  • domain assumption Features within each cluster follow a multivariate Gaussian distribution with diagonal covariance, so KL divergence between Gaussians is a meaningful objective.
    Sec. 3.2 Selection step. This is a heuristic distributional assumption common in FSCIL replay methods, but it is unverified for the feature spaces used.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Class Balance Matters to Active Class-Incremental Learning." pith.science (2026). https://pith.science/paper/XRJIPFGJ

@misc{pith2026241206642,
  author       = {Pith},
  title        = {Pith review of: Class Balance Matters to Active Class-Incremental Learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/XRJIPFGJ}},
  note         = {Machine review of arXiv:2412.06642}
}
read the original abstract

Few-Shot Class-Incremental Learning has shown remarkable efficacy in efficient learning new concepts with limited annotations. Nevertheless, the heuristic few-shot annotations may not always cover the most informative samples, which largely restricts the capability of incremental learner. We aim to start from a pool of large-scale unlabeled data and then annotate the most informative samples for incremental learning. Based on this premise, this paper introduces the Active Class-Incremental Learning (ACIL). The objective of ACIL is to select the most informative samples from the unlabeled pool to effectively train an incremental learner, aiming to maximize the performance of the resulting model. Note that vanilla active learning algorithms suffer from class-imbalanced distribution among annotated samples, which restricts the ability of incremental learning. To achieve both class balance and informativeness in chosen samples, we propose Class-Balanced Selection (CBS) strategy. Specifically, we first cluster the features of all unlabeled images into multiple groups. Then for each cluster, we employ greedy selection strategy to ensure that the Gaussian distribution of the sampled features closely matches the Gaussian distribution of all unlabeled features within the cluster. Our CBS can be plugged and played into those CIL methods which are based on pretrained models with prompts tunning technique. Extensive experiments under ACIL protocol across five diverse datasets demonstrate that CBS outperforms both random selection and other SOTA active learning approaches. Code is publicly available at https://github.com/1170300714/CBS.

Figures

Figures reproduced from arXiv: 2412.06642 by the authors.

Figure 1
Figure 1. Analysis of applying various active learning approaches to LP-DiF [ [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Avg curves of our CBS and comparison with counterparts applied to LP-DiF on five datasets ( [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗
Figure 3
Figure 3. Avg curves of our CBS and comparison with counterparts applied to L2P [ [PITH_FULL_IMAGE:figures/full_fig_p012_3.png] view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: Avg curves of our CBS and comparison with counterparts applied to DualPrompt [ [PITH_FULL_IMAGE:figures/full_fig_p013_4.png]
Figure 5
Figure 5. Figure 5: Comparison of CBS and other counterparts applied to LP-DiF in terms of “class-imbalanced ratio” on CUB-200 under [PITH_FULL_IMAGE:figures/full_fig_p014_5.png]
Figure 6
Figure 6. Figure 6: Comparison of “classes discovery ratio” by our CBS and other counterparts applied to LP-DiF on five datasets ( [PITH_FULL_IMAGE:figures/full_fig_p015_6.png]
Figure 7
Figure 7. Figure 7: The KL divergence between Gaussian distribution estimated by all samples and Gaussian distribution estimated by [PITH_FULL_IMAGE:figures/full_fig_p015_7.png]
Figure 9
Figure 9. Figure 9: Decoupling the performance of the last session to [PITH_FULL_IMAGE:figures/full_fig_p015_9.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

88 extracted references · 45 canonical work pages

  1. [1]

    Touqeer Ahmad, Akshay Raj Dhamija, Steve Cruz, Ryan Rabinowitz, Chunchun Li, Mohsen Jafarzadeh, and Terrance E Boult. 2022. Few-shot class incremen- tal learning leveraging self-supervised features. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition . 3900–3910

    2022

  2. [2]

    Touqeer Ahmad, Akshay Raj Dhamija, Mohsen Jafarzadeh, Steve Cruz, Ryan Rabinowitz, Chunchun Li, and Terrance E Boult. 2022. Variable few shot class incremental and open world learning. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition . 3688–3699

    2022

  3. [3]

    Afra Feyza Akyürek, Ekin Akyürek, Derry Tanti Wijaya, and Jacob Andreas. 2021. Subspace regularizers for few-shot class incremental learning. arXiv preprint arXiv:2110.07059 (2021)

    arXiv 2021

  4. [4]

    Rahaf Aljundi, Punarjay Chakravarty, and Tinne Tuytelaars. 2017. Expert gate: Lifelong learning with a network of experts. In Proceedings of the IEEE conference on computer vision and pattern recognition . 3366–3375

    2017

  5. [5]

    Rahaf Aljundi, Min Lin, Baptiste Goujaud, and Yoshua Bengio. 2019. Gradi- ent based sample selection for online continual learning. Advances in neural information processing systems 32 (2019)

    2019

  6. [6]

    Jordan T Ash, Chicheng Zhang, Akshay Krishnamurthy, John Langford, and Alekh Agarwal. 2019. Deep batch active learning by diverse, uncertain gradient lower bounds. arXiv preprint arXiv:1906.03671 (2019)

    arXiv 2019

  7. [7]

    Jordan T Ash, Chicheng Zhang, Akshay Krishnamurthy, John Langford, and Alekh Agarwal. 2020. Deep Batch Active Learning by Diverse, Uncertain Gradient Lower Bounds. In International Conference on Learning Representations

    2020

  8. [8]

    Ali Ayub and Carter Fendley. 2022. Few-shot continual active learning by a robot. Advances in Neural Information Processing Systems 35 (2022), 30612–30624

    2022

Show all 88 references
  1. [9]

    Jihwan Bang, Heesu Kim, YoungJoon Yoo, Jung-Woo Ha, and Jonghyun Choi

  2. [10]

    Arslan Chaudhry, Puneet K Dokania, Thalaiyasingam Ajanthan, and Philip HS Torr. 2018. Riemannian walk for incremental learning: Understanding forgetting and intransigence. In Proceedings of the European conference on computer vision (ECCV). 532–547

    2018

  3. [11]

    Jinpeng Chen, Runmin Cong, Yuxuan Luo, Horace Ho Shing Ip, and Sam Kwong

  4. [12]

    Ali Cheraghian, Shafin Rahman, Pengfei Fang, Soumava Kumar Roy, Lars Peters- son, and Mehrtash Harandi. 2021. Semantic-aware knowledge distillation for few-shot class-incremental learning. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition . 2534–2543

    2021

  5. [13]

    Ali Cheraghian, Shafin Rahman, Sameera Ramasinghe, Pengfei Fang, Christian Simon, Lars Petersson, and Mehrtash Harandi. 2021. Synthesized feature based few-shot class-incremental learning on a mixture of subspaces. In Proceedings of the IEEE/CVF international conference on com...

    2021

  6. [14]

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

    2022

  7. [15]

    Mircea Cimpoi, Subhransu Maji, Iasonas Kokkinos, Sammy Mohamed, and An- drea Vedaldi. 2014. Describing textures in the wild. In Proceedings of the IEEE conference on computer vision and pattern recognition . 3606–3613

    2014

  8. [16]

    Songlin Dong, Xiaopeng Hong, Xiaoyu Tao, Xinyuan Chang, Xing Wei, and Yihong Gong. 2021. Few-shot class-incremental learning via relation knowledge distillation. In Proceedings of the AAAI Conference on Artificial Intelligence, Vol. 35. 1255–1263

    2021

  9. [17]

    Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xi- aohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, et al. 2020. An image is worth 16x16 words: Transformers for image recognition at scale. arXiv prepri...

    2020 arXiv

  10. [18]

    Arthur Douillard, Alexandre Ramé, Guillaume Couairon, and Matthieu Cord. 2022. Dytox: Transformers for continual learning with dynamic token expansion. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 9285–9295

    2022

  11. [19]

    Yarin Gal, Riashat Islam, and Zoubin Ghahramani. 2017. Deep bayesian active learning with image data. In International conference on machine learning . PMLR, 1183–1192

    2017

  12. [20]

    Qiankun Gao, Chen Zhao, Bernard Ghanem, and Jian Zhang. 2022. R-dfcil: Relation-guided representation learning for data-free class incremental learning. In European Conference on Computer Vision . Springer, 423–439

    2022

  13. [21]

    Ziqi Gu, Chunyan Xu, Jian Yang, and Zhen Cui. 2023. Few-shot Continual Infomax Learning. In Proceedings of the IEEE/CVF International Conference on Computer Vision. 19224–19233

    2023

  14. [22]

    Guy Hacohen, Avihu Dekel, and Daphna Weinshall. 2022. Active Learning on a Budget: Opposite Strategies Suit High and Low Budgets. In International Conference on Machine Learning . PMLR, 8175–8195

    2022

  15. [23]

    Michael Hersche, Geethan Karunaratne, Giovanni Cherubini, Luca Benini, Abu Sebastian, and Abbas Rahimi. 2022. Constrained few-shot class-incremental learning. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 9057–9067

    2022

  16. [24]

    Alex Holub, Pietro Perona, and Michael C Burl. 2008. Entropy-based active learning for object recognition. In 2008 IEEE Computer Society Conference on Computer Vision and Pattern Recognition Workshops . IEEE, 1–8

    2008

  17. [25]

    Saihui Hou, Xinyu Pan, Chen Change Loy, Zilei Wang, and Dahua Lin. 2019. Learning a unified classifier incrementally via rebalancing. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition . 831–839

    2019

  18. [26]

    Zitong Huang, Ze Chen, Zhixing Chen, Erjin Zhou, Xinxing Xu, Rick Siow Mong Goh, Yong Liu, Chunmei Feng, and Wangmeng Zuo. 2024. Learning Prompt with Distribution-Based Feature Replay for Few-Shot Class-Incremental Learning. arXiv preprint arXiv:2401.01598 (2024)

    2024 arXiv

  19. [27]

    David Isele and Akansel Cosgun. 2018. Selective experience replay for lifelong learning. In Proceedings of the AAAI Conference on Artificial Intelligence , Vol. 32

    2018

  20. [28]

    Menglin Jia, Luming Tang, Bor-Chun Chen, Claire Cardie, Serge Belongie, Bharath Hariharan, and Ser-Nam Lim. 2022. Visual prompt tuning. In Euro- pean Conference on Computer Vision . Springer, 709–727

    2022

  21. [29]

    Do-Yeon Kim, Dong-Jun Han, Jun Seo, and Jaekyun Moon. 2022. Warping the space: Weight space rotation for class-incremental few-shot learning. In The Eleventh International Conference on Learning Representations

    2022

  22. [30]

    James Kirkpatrick, Razvan Pascanu, Neil Rabinowitz, Joel Veness, Guillaume Desjardins, Andrei A Rusu, Kieran Milan, John Quan, Tiago Ramalho, Agnieszka Grabska-Barwinska, et al. 2017. Overcoming catastrophic forgetting in neural networks. Proceedings of the national academy of...

    2017

  23. [31]

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

    2009

  24. [32]

    Anna Kukleva, Hilde Kuehne, and Bernt Schiele. 2021. Generalized and incre- mental few-shot learning by explicit learning and calibration without forgetting. In Proceedings of the IEEE/CVF international conference on computer vision . 9020– 9029

    2021

  25. [33]

    Sang-Woo Lee, Jin-Hwa Kim, Jaehyun Jun, Jung-Woo Ha, and Byoung-Tak Zhang

  26. [34]

    Zhizhong Li and Derek Hoiem. 2017. Learning without forgetting. IEEE transac- tions on pattern analysis and machine intelligence 40, 12 (2017), 2935–2947

    2017

  27. [35]

    Yuxuan Luo, Runmin Cong, Xialei Liu, Horace Ho Shing Ip, and Sam Kwong

  28. [36]

    James MacQueen et al. 1967. Some methods for classification and analysis of multivariate observations. In Proceedings of the fifth Berkeley symposium on mathematical statistics and probability , Vol. 1. Oakland, CA, USA, 281–297. MM ’24, October 28–November 1, 2024, Melbourne,...

    1967

  29. [37]

    Pratik Mazumder, Pravendra Singh, and Piyush Rai. 2021. Few-shot lifelong learning. In Proceedings of the AAAI Conference on Artificial Intelligence , Vol. 35. 2337–2345

    2021

  30. [38]

    Maria-Elena Nilsback and Andrew Zisserman. 2008. Automated flower classifica- tion over a large number of classes. In 2008 Sixth Indian conference on computer vision, graphics & image processing . IEEE, 722–729

    2008

  31. [39]

    IEEE Transactions on Multimedia (2024)

    Modeling Inner-and Cross-Task Contrastive Relations for Continual Image Classification. IEEE Transactions on Multimedia (2024)

    2024

  32. [40]

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. 2021. Learning transferable visual models from natural language supervision. In International conference on machine learni...

    2021

  33. [41]

    Sanket Rajan Gupte, Josiah Aklilu, Jeffrey J Nirschl, and Serena Yeung-Levy. 2024. Revisiting Active Learning in the Era of Vision Foundation Models.arXiv e-prints (2024), arXiv–2401

    2024

  34. [42]

    Sylvestre-Alvise Rebuffi, Alexander Kolesnikov, Georg Sperl, and Christoph H Lampert. 2017. icarl: Incremental classifier and representation learning. In Pro- ceedings of the IEEE conference on Computer Vision and Pattern Recognition . 2001– 2010

    2017

  35. [43]

    Amin Parvaneh, Ehsan Abbasnejad, Damien Teney, Gholamreza Reza Haffari, Anton Van Den Hengel, and Javen Qinfeng Shi. 2022. Active learning by feature mixing. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 12237–12246

    2022

  36. [44]

    Olga Russakovsky, Jia Deng, Hao Su, Jonathan Krause, Sanjeev Satheesh, Sean Ma, Zhiheng Huang, Andrej Karpathy, Aditya Khosla, Michael Bernstein, et al

  37. [45]

    Andrei A Rusu, Neil C Rabinowitz, Guillaume Desjardins, Hubert Soyer, James Kirkpatrick, Koray Kavukcuoglu, Razvan Pascanu, and Raia Hadsell. 2016. Pro- gressive neural networks. arXiv preprint arXiv:1606.04671 (2016)

    2016 arXiv

  38. [46]

    Ozan Sener and Silvio Savarese. 2017. Active learning for convolutional neural networks: A core-set approach. arXiv preprint arXiv:1708.00489 (2017)

    2017 arXiv

  39. [47]

    Dan Roth and Kevin Small. 2006. Margin-based active learning for structured output spaces. In Machine Learning: ECML 2006: 17th European Conference on Machine Learning Berlin, Germany, September 18-22, 2006 Proceedings 17 . Springer, 413–424

    2006

  40. [48]

    James Seale Smith, Leonid Karlinsky, Vyshnavi Gutta, Paola Cascante-Bonilla, Donghyun Kim, Assaf Arbelle, Rameswar Panda, Rogerio Feris, and Zsolt Kira

  41. [49]

    Xiaoyu Tao, Xiaopeng Hong, Xinyuan Chang, Songlin Dong, Xing Wei, and Yihong Gong. 2020. Few-shot class-incremental learning. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition . 12183–12192

    2020

  42. [50]

    Vishal Thengane, Salman Khan, Munawar Hayat, and Fahad Khan. 2022. Clip model is an efficient continual learner. arXiv preprint arXiv:2210.03114 (2022)

    2022 arXiv

  43. [51]

    Songsong Tian, Lusi Li, Weijun Li, Hang Ran, Xin Ning, and Prayag Tiwari. 2023. A survey on few-shot class-incremental learning. arXiv preprint arXiv:2304.08130 (2023)

    2023 arXiv

  44. [52]

    Ozan Sener and Silvio Savarese. 2018. Active Learning for Convolutional Neu- ral Networks: A Core-Set Approach. In International Conference on Learning Representations

    2018

  45. [53]

    Runqi Wang, Xiaoyue Duan, Guoliang Kang, Jianzhuang Liu, Shaohui Lin, Song- cen Xu, Jinhu Lü, and Baochang Zhang. 2023. AttriCLIP: A Non-Incremental Learner for Incremental Knowledge Learning. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition...

    2023

  46. [54]

    Yabin Wang, Zhiwu Huang, and Xiaopeng Hong. 2022. S-prompts learning with pre-trained transformers: An occam’s razor for domain incremental learning. Advances in Neural Information Processing Systems 35 (2022), 5682–5695

    2022

  47. [55]

    Zifeng Wang, Zizhao Zhang, Sayna Ebrahimi, Ruoxi Sun, Han Zhang, Chen-Yu Lee, Xiaoqi Ren, Guolong Su, Vincent Perot, Jennifer Dy, et al. 2022. Dualprompt: Complementary prompting for rehearsal-free continual learning. In European Conference on Computer Vision . Springer, 631–648

    2022

  48. [56]

    Zifeng Wang, Zizhao Zhang, Chen-Yu Lee, Han Zhang, Ruoxi Sun, Xiaoqi Ren, Guolong Su, Vincent Perot, Jennifer Dy, and Tomas Pfister. 2022. Learning to prompt for continual learning. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition . 139–149

    2022

  49. [57]

    Yichen Xie, Mingyu Ding, Masayoshi Tomizuka, and Wei Zhan. 2024. Towards free data selection with general-purpose models. Advances in Neural Information Processing Systems 36 (2024)

    2024

  50. [58]

    Catherine Wah, Steve Branson, Peter Welinder, Pietro Perona, and Serge Belongie

  51. [59]

    Shipeng Yan, Jiangwei Xie, and Xuming He. 2021. Der: Dynamically expandable representation for class incremental learning. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition . 3014–3023

    2021

  52. [60]

    Boyu Yang, Mingbao Lin, Binghao Liu, Mengying Fu, Chang Liu, Rongrong Ji, and Qixiang Ye. 2021. Learnable Expansion-and-Compression Network for Few-shot Class-Incremental Learning. arXiv preprint arXiv:2104.02281 (2021)

    2021 arXiv

  53. [61]

    Boyu Yang, Mingbao Lin, Yunxiao Zhang, Binghao Liu, Xiaodan Liang, Rongrong Ji, and Qixiang Ye. 2022. Dynamic support network for few-shot class incremental learning. IEEE Transactions on Pattern Analysis and Machine Intelligence 45, 3 (2022), 2945–2951

    2022

  54. [62]

    Yang Yang, Zhiying Cui, Junjie Xu, Changhong Zhong, Wei-Shi Zheng, and Ruixuan Wang. 2023. Continual learning with Bayesian model based on a fixed pre-trained feature extractor. Visual Intelligence 1, 1 (2023), 5

    2023

  55. [63]

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

    2023 arXiv

  56. [64]

    Yang Yang, Da-Wei Zhou, De-Chuan Zhan, Hui Xiong, and Yuan Jiang. 2019. Adaptive deep models for incremental learning: Considering capacity scalability and sustainability. In Proceedings of the 25th ACM SIGKDD International Confer- ence on Knowledge Discovery & Data Mining . 74–82

    2019

  57. [65]

    Zhengzhuo Xu, Ruikang Liu, Shuo Yang, Zenghao Chai, and Chun Yuan. 2023. Learning Imbalanced Data with Vision Transformers. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition . 15793–15803

    2023

  58. [66]

    Jaehong Yoon, Eunho Yang, Jeongtae Lee, and Sung Ju Hwang. 2017. Lifelong learning with dynamically expandable networks. arXiv preprint arXiv:1708.01547 (2017)

    2017 arXiv

  59. [67]

    Friedemann Zenke, Ben Poole, and Surya Ganguli. 2017. Continual learning through synaptic intelligence. In International conference on machine learning . PMLR, 3987–3995

    2017

  60. [68]

    Xueying Zhan, Qingzhong Wang, Kuan-hao Huang, Haoyi Xiong, Dejing Dou, and Antoni B Chan. 2022. A comparative survey of deep active learning. arXiv preprint arXiv:2203.13450 (2022)

    2022 arXiv

  61. [69]

    Chi Zhang, Nan Song, Guosheng Lin, Yun Zheng, Pan Pan, and Yinghui Xu

  62. [70]

    Yabo Zhang, Yuxiang Wei, Dongsheng Jiang, Xiaopeng Zhang, Wangmeng Zuo, and Qi Tian. 2023. Controlvideo: Training-free controllable text-to-video genera- tion. arXiv preprint arXiv:2305.13077 (2023)

    2023 arXiv

  63. [71]

    Hanbin Zhao, Yongjian Fu, Mintong Kang, Qi Tian, Fei Wu, and Xi Li. 2021. Mgsvf: Multi-grained slow vs. fast framework for few-shot class-incremental learning. IEEE Transactions on Pattern Analysis and Machine Intelligence (2021)

    2021

  64. [72]

    Ofer Yehuda, Avihu Dekel, Guy Hacohen, and Daphna Weinshall. 2022. Active learning through a covering lens. Advances in Neural Information Processing Systems 35 (2022), 22354–22367

    2022

  65. [73]

    Da-Wei Zhou, Qi-Wei Wang, Zhi-Hong Qi, Han-Jia Ye, De-Chuan Zhan, and Ziwei Liu. 2023. Deep class-incremental learning: A survey. arXiv preprint arXiv:2302.03648 (2023)

    2023 arXiv

  66. [74]

    Da-Wei Zhou, Han-Jia Ye, Liang Ma, Di Xie, Shiliang Pu, and De-Chuan Zhan

  67. [75]

    Kaiyang Zhou, Jingkang Yang, Chen Change Loy, and Ziwei Liu. 2022. Learning to prompt for vision-language models. International Journal of Computer Vision 130, 9 (2022), 2337–2348

    2022

  68. [76]

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

    2021

  69. [77]

    In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Few-shot incremental learning with continually evolved classifiers. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition . 12455–12464

  70. [78]

    Yixiong Zou, Shanghang Zhang, Yuhua Li, and Ruixuan Li. 2022. Margin-based few-shot class-incremental learning with class-level overfitting mitigation. Ad- vances in neural information processing systems 35 (2022), 27267–27279. Class Balance Matters to Active Class-Incremental...

    2022

  71. [80]

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

    2022

  72. [86]

    Huiping Zhuang, Zhenyu Weng, Run He, Zhiping Lin, and Ziqian Zeng. 2023. GKEAL: Gaussian Kernel Embedded Analytic Learning for Few-Shot Class Incre- mental Task. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 7746–7755

    2023

  73. [88]

    class- imbalanced ratio

    For both L2P and DualPrompt, our CBS achieves the highest Mean Avg over five datasets under each labeling budget compared to all the counterparts. The above results, along with those of LP-DiF in the main paper, fully demonstrate that our CBS can be plug-and- played with these...

    2024

  74. [2011]

    The caltech-ucsd birds-200-2011 dataset. (2011)

    2011

  75. [2015]

    International journal of computer vision 115 (2015), 211–252

    Imagenet large scale visual recognition challenge. International journal of computer vision 115 (2015), 211–252

    2015

  76. [2017]

    Advances in neural information processing systems 30 (2017)

    Overcoming catastrophic forgetting by incremental moment matching. Advances in neural information processing systems 30 (2017)

    2017

  77. [2021]

    In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Rainbow memory: Continual learning with a memory of diverse samples. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition . 8218–8227

  78. [2022]

    IEEE Transactions on Pattern Analysis and Machine Intelligence (2022)

    Few-shot class-incremental learning by sampling multi-phase tasks. IEEE Transactions on Pattern Analysis and Machine Intelligence (2022)

    2022

  79. [2023]

    In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    CODA-Prompt: COntinual Decomposed Attention-based Prompting for Rehearsal-Free Continual Learning. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition . 11909–11919

  80. [2024]

    arXiv preprint arXiv:2407.16354 (2024)

    Strike a Balance in Continual Panoptic Segmentation. arXiv preprint arXiv:2407.16354 (2024)

    2024 arXiv

Pith tools

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