Pith. sign in

REVIEW 4 major objections 5 minor 57 references

Replacing human labels with confidence-filtered KMeans pseudo-labels lets standard class-incremental learning methods train on unlabeled datasets, losing on average about 10 points of final accuracy and beating adapted novel-category-discov

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

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

ICPL generates pseudo-labels by clustering embeddings with KMeans, keeps only confident ones, and uses them to train class-incremental models without human labels, losing about 10 points versus supervised CIL but beating class-iNCD baselines by more than 5.

T0 review reviewed 2026-08-05 challenge →

load-bearing objection A straightforward pseudo-label pipeline with a genuinely useful evaluation protocol, undermined by an over-broad 'unsupervised' claim and one-seed results. the 4 major comments →

arxiv 2508.21424 v2 pith:T572PYZP submitted 2025-08-29 cs.CV

Unsupervised Incremental Learning Using Confidence-Based Pseudo-Labels

Lucas Rakotoarivony This is my paper
classification cs.CV
keywords unsupervised class-incremental learningpseudo-labelsKMeans clusteringconfidence selectioncatastrophic forgettingnovel category discoveryCIFAR100ImageNet100
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

Standard class-incremental learning assumes every new batch of classes arrives with human labels. This paper argues that the labels can be replaced by pseudo-labels generated from KMeans clustering of the model's own feature embeddings, with a confidence threshold discarding the least reliable clusters, and that this cost saving costs surprisingly little accuracy. On CIFAR100 and ImageNet100, plugging these pseudo-labels into four established CIL methods (Replay, iCaRL, WA, FOSTER) yields final accuracies on average about 10 points below the same methods with true labels, and in one configuration the drop is only 1.33 points. Against adapted class-iNCD and generalized category discovery baselines designed for the same unlabeled setting, ICPL is more than 5 points better in final accuracy on CIFAR100, especially when many incremental tasks are run. The paper also introduces a static encoding evaluation protocol that penalizes confusion between old and new classes, which cluster-accuracy metrics miss.

Core claim

ICPL's central claim is that the bottleneck for real-world incremental learning is not the learning algorithm but the annotation pipeline: given an unlabeled batch of a known number of new classes, KMeans on the embeddings of the previous model produces clusters that can stand in for ground-truth labels, and a softmax over inverse KMeans distances gives a confidence score that lets the method keep only the most reliable pseudo-labels (threshold α=0.85, recomputed every 10 epochs). Combining these pseudo-labels with off-the-shelf CIL methods—substituting ŷ for y—keeps the model within roughly 10 points of supervised accuracy on average, outperforms class-iNCD methods adapted to long task sequ

What carries the argument

KMeans clustering over the feature embeddings of the unlabeled new-class data, which assigns each image a pseudo-label from the cluster index; a confidence score for each pseudo-label computed as a softmax over the inverse of the KMeans distance matrix (so samples far from all cluster centers get low confidence), with a threshold α=0.85 selecting a subset for training; and a static encoding evaluation protocol in which the Hungarian assignment maps pseudo-classes to ground-truth classes once per task and is frozen, so later tasks cannot rewrite earlier class identities. The pseudo-labels are fed into standard CIL methods (Replay, iCaRL, WA, FOSTER) in place of human labels; periodic recomput

Load-bearing premise

The method only works if the number of new classes in each incremental task is known in advance, since KMeans is given that number as its cluster count; if the true class count is unknown, the pseudo-label generation cannot be applied as described.

What would settle it

Run the WA+ICPL pipeline on CIFAR100 Base50 Inc10 with the per-task cluster count set to n−1, n, n+1 (and to n/2) and compare final accuracy; if the method tolerates wrong cluster counts, its stated dependence on known n is not the load-bearing constraint. Alternatively, on a held-in incremental task, compare the ground-truth accuracy of pseudo-labels above the α=0.85 threshold against those below it; if high-confidence pseudo-labels are not systematically more accurate, the confidence selection is not doing the work claimed.

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

If this is right

  • Any existing CIL method can be made label-free by swapping human labels for ICPL pseudo-labels, so the annotation cost of continual learning in new domains drops by the entire labeling budget of the incremental tasks.
  • The average final-accuracy gap to supervised CIL of about 10 points, shrinking to 1.33 points in the CIFAR100 Base50 Inc10 configuration with FOSTER, defines what label noise costs in practice; methods that improve embedding separability should narrow it further.
  • ICPL's more-than-5-point final-accuracy advantage over adapted FRoST, GM, DCCL, and CMS on CIFAR100 indicates that long task sequences are a regime class-iNCD methods currently do not handle well, while CIL-plus-pseudo-labels does.
  • Because low-confidence samples are dropped (about 30% at α=0.85), ICPL training uses roughly 24% fewer TFLOPs than supervised training, making unlabeled continual learning cheaper than labeled continual learning.
  • The static-encoding protocol, not cluster accuracy, is the right way to measure success in this setting; cluster accuracy masks cross-task confusion that a deployed model would exhibit.

Where Pith is reading between the lines

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

  • ICPL could serve as a probe decomposing a CIL method's failure into forgetting versus labeling noise: run the method supervised and unsupervised on the same stream; the gap isolates the effect of imperfect pseudo-labels, and the residual drop after correcting labels isolates forgetting.
  • Because the known-n assumption is the fragile part, an immediate extension is estimating k per task; if k is over- or under-specified, the pseudo-label quality and thus the 10-point margin will likely degrade, so methods to estimate class counts could make it fully unsupervised.
  • The confidence selection's utility implies that embeddings from the previous task's model must separate new classes well; applying stronger representation learning (e.g., contrastive objectives on the unlabeled data before clustering) could push the unsupervised numbers above the 5-point margin or even close the gap to supervised.
  • The static encoding protocol is reusable beyond this paper; if adopted as a standard metric for unsupervised CIL, it would expose cross-task confusion in other methods that currently report cluster accuracy, likely changing published conclusions.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper proposes ICPL, an unsupervised class-incremental learning pipeline that replaces human annotations for incremental tasks with KMeans-based pseudo-labels. For each new task, embeddings are extracted with the current model, KMeans is applied with the (assumed known) number of classes, pseudo-labels are assigned by nearest cluster, and a confidence threshold derived from softmax-transformed cluster distances selects a subset of samples. These pseudo-labels are plugged into four standard CIL methods (Replay, iCaRL, WA, FOSTER). The paper also introduces a 'static encoding' evaluation protocol that uses a fixed Hungarian assignment between pseudo-classes and ground-truth classes. Experiments on CIFAR100, ImageNet100, and fine-grained datasets report that ICPL is within about 10 points of supervised CIL (sometimes 1.33 points) and outperforms adapted class-iNCD/GCD baselines by more than 5% final accuracy, while reducing training cost.

Significance. If the results are reproducible, ICPL would be a useful, simple baseline for unsupervised CIL, and the static-encoding protocol is a genuine improvement over raw cluster accuracy for measuring cross-task confusion. The paper is transparent about the known-class-count assumption in Section 3 and provides extensive ablations over the confidence threshold, pseudo-label recomputation step, clustering algorithm, data augmentation, and per-method modifications. However, the strong comparative claims (e.g., >5% over class-iNCD, 1.33% gap to supervised CIL) rest on single runs and on hyperparameters selected using the same benchmarks, so the significance is conditional on the concerns below being addressed.

major comments (4)
  1. [Section 3, Algorithm 1] The method is only defined when the exact number of classes in every new task is supplied: Section 3 states 'n is assumed to be known in our unsupervised setting', and Algorithm 1 requires the number of clusters to be exactly n. The abstract and contribution list say the method 'replaces human annotations' and 'enables incremental learning from unlabeled datasets' without this caveat. Exact class count is a form of supervision that is not available in many real streams. Please either relax the assumption (e.g., estimate the number of clusters or evaluate sensitivity to misspecified k) or state the assumption prominently in the title/abstract.
  2. [Table 9 vs Tables 1 and 7] There is a numerical inconsistency for the same configuration. Table 9 reports, for WA+ICPL with Base0 Inc20 and alpha=0.85, final/mean accuracy 37.78/53.52 and 73% skipped samples, while Table 1 and Table 7 (tau=10, same setting) report 36.78/53.22. This is not a formatting difference: the default-alpha setting in the main results disagrees with the ablation table. Please correct the inconsistency and rerun/verify the affected numbers.
  3. [Section 4.3, Appendix B] The choice to add MixUp and class-weight balancing only for Replay and iCaRL was made after observing performance on the same CIFAR100 configurations, as stated in Section 4.3 ('we experimentally found that these methods can improve performance') and shown in Tables 8 and 10. These choices change final accuracy by several points, so the reported comparison among ICPL variants is partly a result of model selection on the evaluation benchmarks. At minimum, disclose this selection procedure; better, validate the choice on an independent split or with a pre-registered protocol.
  4. [Tables 1, 2, 4, 7] All quantitative claims are based on a single run with no error bars or multiple seeds. The headline margins (e.g., >5% over class-iNCD, 1.33% gap to supervised) are comparable to or smaller than the seed-to-seed variability typically observed for CIFAR100 CIL. Please report at least 3 seeds with mean and standard deviation for the main tables, and provide a significance statement for the central comparative claims.
minor comments (5)
  1. [Section 3, Algorithm 1] Notation is overloaded: Section 3 defines n as the number of training instances, while Base-m Inc-n and Section 4.2 use n as the number of classes per task, and Algorithm 1 does not list the number of clusters as an input. Please use separate symbols (e.g., N_inst and C) consistently.
  2. [Appendix C, Equations (2)-(4)] The arithmetic in the GFLOPs calculation is inconsistent. From the stated values, Eq. (2) gives 0.14*5000 + 0.16 = 700.16 GFLOPs, not 688.66; Eq. (4) does not evaluate to 360207 GFLOPs; and the later '11707 GFLOPs' figure is not derived. Please recompute and correct the equations and the reported percentage reduction.
  3. [Table 9] The claim that 'the default threshold of 0.85 provides the best overall performance' is not supported for several individual configurations: in Base0 Inc5, alpha=0.65 and 0.75 give higher final accuracy; in Base50 Inc10, alpha=0.95 gives higher final accuracy. State the aggregation criterion (e.g., mean average accuracy over settings) and report per-configuration variance.
  4. [Section 6.2] The sentence 'the differences are small... suggesting this issue may be optional' is contradicted by the following paragraph arguing that the cluster-accuracy protocol is dangerous. Rephrase to avoid ambiguity about whether the authors believe the evaluation protocol change is necessary.
  5. [Table 1 / Figure 4] Table 1 uses 'Base0'/'B50' while Figure 4 uses 'Base10' in the ImageNet100 panel; also the method name 'W A' is written with a space. Please make labels consistent.

Circularity Check

0 steps flagged

No construction-level circularity: ICPL is an empirical pipeline whose pseudo-labels come from KMeans on unlabeled embeddings, and the reported comparisons do not reduce to the method's own inputs.

full rationale

The paper makes no derivation claim that collapses into its inputs. Algorithm 1 generates pseudo-labels from KMeans on feature embeddings, with confidence computed from the same distance matrix; this is a self-contained unsupervised labeling step, and the CIL methods (Replay, iCaRL, WA, FOSTER) are externally defined baselines into which the pseudo-labels are plugged. The static-encoding evaluation protocol uses Hungarian assignment to ground truth, but the paper explicitly states labels are used only for testing (Section 4.1), and Section 6.2 shows the protocol changes results only slightly. The main caveats are non-circular: (i) Section 3 states the number of classes per incremental task is assumed known, which is a supervision assumption rather than a circular one; (ii) hyperparameters alpha=0.85, tau=10, AutoAugment and MixUp choices are selected via ablations on the same CIFAR100 benchmark, which is a benchmark-tuning/overfitting concern, not a construction-level circularity; (iii) there are no load-bearing self-citations by the author. Therefore no step in the central claim is equivalent to its inputs by definition.

Axiom & Free-Parameter Ledger

2 free parameters · 4 axioms · 0 invented entities

The method introduces no new physical or model entities; it relies on known clustering and pseudo-labeling machinery. The main assumptions are the known class count, clusterability of embeddings, and validity of the static encoding protocol.

free parameters (2)
  • Confidence threshold alpha = 0.85
    Chosen by ablation on CIFAR100 (Table 9); it controls the tradeoff between pseudo-label quality and quantity, and the paper reports performance varies with it.
  • Pseudo-label recomputation step size tau = 10 epochs
    Selected by ablation (Table 7); larger or smaller tau degrade results, and this is tuned on the same benchmark.
axioms (4)
  • domain assumption The number of new classes n is known for every incremental task.
    Stated in Section 3. KMeans needs k=n to form clusters that correspond to classes.
  • domain assumption Feature embeddings of images from the same new class are closer to each other than to other classes, so KMeans clusters recover class structure.
    Central to pseudo-label quality; invoked in Section 4.2 with citation [3,7].
  • domain assumption The static encoding evaluation protocol, which assigns cluster-to-class identities via Hungarian matching and fixes them over time, is a valid measure of incremental classification accuracy.
    Proposed in Section 4.1; all headline comparisons use it, and it is stricter than cluster accuracy but not a universally accepted benchmark.
  • standard math KMeans converges to a reasonable local optimum on deep embeddings; no guarantee is given.
    Standard behavior of KMeans as used in Algorithm 1.

reviewed 2026-08-05 · how reviews work

0 comments
Cite this review

Pith. "Pith review of Unsupervised Incremental Learning Using Confidence-Based Pseudo-Labels." pith.science (2026). https://pith.science/paper/T572PYZP

@misc{pith2026250821424,
  author       = {Pith},
  title        = {Pith review of: Unsupervised Incremental Learning Using Confidence-Based Pseudo-Labels},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/T572PYZP}},
  note         = {Machine review of arXiv:2508.21424}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Deep learning models have achieved state-of-the-art performance in many computer vision tasks. However, in real-world scenarios, novel classes that were unseen during training often emerge, requiring models to acquire new knowledge incrementally. Class-Incremental Learning (CIL) methods enable a model to learn novel classes while retaining knowledge of previous classes. However, these methods make the strong assumption that the incremental dataset is fully labeled, which is unrealistic in practice. In this work, we propose an unsupervised Incremental Learning method using Confidence-based Pseudo-labels (ICPL), which replaces human annotations with pseudo-labels, enabling incremental learning from unlabeled datasets. We integrate these pseudo-labels into various CIL methods with confidence-based selection and evaluate performance degradation on CIFAR100 and ImageNet100. Then, we compare our approach to popular Class Incremental Novel Category Discovery (class-iNCD) methods addressing similar challenges. Additionally, we apply our method to fine-grained datasets to demonstrate its real-world practicality and measure its computational complexity to validate its suitability for resource-constrained environments. ICPL achieves competitive results compared to supervised methods and outperforms state-of-the-art class-iNCD methods by more than 5% in final accuracy.

Figures

Figures reproduced from arXiv: 2508.21424 by Lucas Rakotoarivony.

Figure 1
Figure 1. Figure 1: Comparison of (a) supervised CIL, which uses a labeled [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Comparison of evaluation protocol (a) Cluster accuracy evaluation, which obtains 100% of accuracy while confusing old and [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Illustration of the generation of pseudo-labels. Firstly, [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Incremental accuracy curves for supervised and unsupervised cases on CIFAR100 and ImageNet100 datasets. [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: An overview of ICPL. A. Overview [PITH_FULL_IMAGE:figures/full_fig_p012_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Graph depicting the evolution of NMI and ARI between selected pseudo-labels and ground-truth labels over epochs, for various [PITH_FULL_IMAGE:figures/full_fig_p015_6.png] view at source ↗

discussion (0)

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

Reference graph

Works this paper leans on

57 extracted references · 46 canonical work pages · 3 internal anchors

  1. [1]

    Rainbow memory: Continual learn- ing with a memory of diverse samples

    Jihwan Bang, Heesu Kim, YoungJoon Yoo, Jung-Woo Ha, and Jonghyun Choi. Rainbow memory: Continual learn- ing with a memory of diverse samples. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 8218–8227, 2021. 2

    work page 2021

  2. [2]

    Scail: Classifier weights scaling for class incremental learning

    Eden Belouadah and Adrian Popescu. Scail: Classifier weights scaling for class incremental learning. In Proceed- ings of the IEEE/CVF winter conference on applications of computer vision, pages 1266–1275, 2020. 13

    work page 2020

  3. [3]

    Deep clustering for unsupervised learning of visual features

    Mathilde Caron, Piotr Bojanowski, Armand Joulin, and Matthijs Douze. Deep clustering for unsupervised learning of visual features. In Proceedings of the European confer- ence on computer vision (ECCV) , pages 132–149, 2018. 3, 4

    work page 2018

  4. [4]

    Riemannian walk for incremen- tal learning: Understanding forgetting and intransigence

    Arslan Chaudhry, Puneet K Dokania, Thalaiyasingam Ajan- than, and Philip HS Torr. Riemannian walk for incremen- tal learning: Understanding forgetting and intransigence. In Proceedings of the European conference on computer vision (ECCV), pages 532–547, 2018. 2

    2018

  5. [5]

    Contrastive mean- shift learning for generalized category discovery

    Sua Choi, Dahyun Kang, and Minsu Cho. Contrastive mean- shift learning for generalized category discovery. In Pro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 23094–23104, 2024. 6, 7

    work page 2024

  6. [6]

    Autoaugment: Learning augmentation policies from data

    Ekin D Cubuk, Barret Zoph, Dandelion Mane, Vijay Vasude- van, and Quoc V Le. Autoaugment: Learning augmentation policies from data. arXiv preprint arXiv:1805.09501, 2018. 5, 12

    Pith/arXiv arXiv 2018

  7. [7]

    A survey on network embedding

    Peng Cui, Xiao Wang, Jian Pei, and Wenwu Zhu. A survey on network embedding. IEEE transactions on knowledge and data engineering, 31(5):833–852, 2018. 4

    work page 2018

  8. [8]

    Dytox: Transformers for continual learning with dynamic token expansion

    Arthur Douillard, Alexandre Ram ´e, Guillaume Couairon, and Matthieu Cord. Dytox: Transformers for continual learning with dynamic token expansion. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 9285–9295, 2022. 2

    work page 2022

  9. [9]

    Xcon: Learning with experts for fine-grained category dis- covery

    Yixin Fei, Zhongkai Zhao, Siwei Yang, and Bingchen Zhao. Xcon: Learning with experts for fine-grained category dis- covery. arXiv preprint arXiv:2208.01898, 2022. 2

    Pith/arXiv arXiv 2022

  10. [10]

    Quick-means: accelerating inference for k-means by learning fast transforms

    Luc Giffon, Valentin Emiya, Hachem Kadri, and Liva Ralaivola. Quick-means: accelerating inference for k-means by learning fast transforms. Machine Learning, 110(5):881– 905, 2021. 14

    work page 2021

  11. [11]

    Automatically Discovering and Learning New Visual Categories with Ranking Statistics

    Kai Han, Sylvestre-Alvise Rebuffi, Sebastien Ehrhardt, An- drea Vedaldi, and Andrew Zisserman. Automatically discov- ering and learning new visual categories with ranking statis- tics. arXiv preprint arXiv:2002.05714, 2020. 2

    work page internal anchor Pith review Pith/arXiv arXiv 2002

  12. [12]

    Autonovel: Automati- cally discovering and learning novel visual categories

    Kai Han, Sylvestre-Alvise Rebuffi, Sebastien Ehrhardt, An- drea Vedaldi, and Andrew Zisserman. Autonovel: Automati- cally discovering and learning novel visual categories. IEEE Transactions on Pattern Analysis and Machine Intelligence, 44(10):6767–6781, 2021. 2

    work page 2021

  13. [13]

    Unsupervised contin- ual learning via pseudo labels

    Jiangpeng He and Fengqing Zhu. Unsupervised contin- ual learning via pseudo labels. In International Work- shop on Continual Semi-Supervised Learning, pages 15–32. Springer, 2021. 1, 3, 5

    work page 2021

  14. [14]

    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. 5

    work page 2016

  15. [15]

    Rethinking im- agenet pre-training

    Kaiming He, Ross Girshick, and Piotr Doll´ar. Rethinking im- agenet pre-training. In Proceedings of the IEEE/CVF inter- national conference on computer vision , pages 4918–4927,

    work page

  16. [16]

    Distill- ing the knowledge in a neural network

    Geoffrey Hinton, Oriol Vinyals, and Jeff Dean. Distill- ing the knowledge in a neural network. arXiv preprint arXiv:1503.02531, 2015. 2

    Pith/arXiv arXiv 2015

  17. [17]

    Learning a unified classifier incrementally via rebalancing

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

    work page

  18. [18]

    Cal- ibrated neighborhood aware confidence measure for deep metric learning

    Maryna Karpusha, Sunghee Yun, and Istvan Fehervari. Cal- ibrated neighborhood aware confidence measure for deep metric learning. arXiv preprint arXiv:2006.04935, 2020. 2, 4

    Pith/arXiv arXiv 2006

  19. [19]

    Unsupervised Class-Incremental Learning Through Confusion

    Shivam Khare, Kun Cao, and James Rehg. Unsupervised class-incremental learning through confusion. arXiv preprint arXiv:2104.04450, 2021. 1

    work page internal anchor Pith review Pith/arXiv arXiv 2021

  20. [20]

    Proxy anchor-based unsu- pervised learning for continuous generalized category dis- covery

    Hyungmin Kim, Sungho Suh, Daehwan Kim, Daun Jeong, Hansang Cho, and Junmo Kim. Proxy anchor-based unsu- pervised learning for continuous generalized category dis- covery. In Proceedings of the IEEE/CVF International Con- ference on Computer Vision, pages 16688–16697, 2023. 5, 7

    work page 2023

  21. [21]

    3d object representations for fine-grained categorization

    Jonathan Krause, Michael Stark, Jia Deng, and Li Fei-Fei. 3d object representations for fine-grained categorization. In Proceedings of the IEEE international conference on com- puter vision workshops, pages 554–561, 2013. 7

    work page 2013

  22. [22]

    Learning multiple layers of features from tiny images

    Alex Krizhevsky, Geoffrey Hinton, et al. Learning multiple layers of features from tiny images. 2009. 2, 5

    work page 2009

  23. [23]

    The hungarian method for the assignment problem

    Harold W Kuhn. The hungarian method for the assignment problem. Naval research logistics quarterly , 2(1-2):83–97,

  24. [24]

    Pseudo-label: The simple and effi- cient semi-supervised learning method for deep neural net- works

    Dong-Hyun Lee et al. Pseudo-label: The simple and effi- cient semi-supervised learning method for deep neural net- works. In Workshop on challenges in representation learn- ing, ICML, page 896. Atlanta, 2013. 2

    work page 2013

  25. [25]

    Large scale k- means clustering using gpus

    Mi Li, Eibe Frank, and Bernhard Pfahringer. Large scale k- means clustering using gpus. Data Mining and Knowledge Discovery, 37(1):67–109, 2023. 8

    work page 2023

  26. [26]

    Learning without forgetting

    Zhizhong Li and Derek Hoiem. Learning without forgetting. IEEE transactions on pattern analysis and machine intelli- gence, 40(12):2935–2947, 2017. 1, 2

    2017

  27. [27]

    Least squares quantization in pcm

    Stuart Lloyd. Least squares quantization in pcm. IEEE trans- actions on information theory, 28(2):129–137, 1982. 2, 3, 4

    work page 1982

  28. [28]

    S. Maji, J. Kannala, E. Rahtu, M. Blaschko, and A. Vedaldi. Fine-grained visual classification of aircraft. Technical re- port, 2013. 7

    work page 2013

  29. [29]

    Catastrophic inter- ference in connectionist networks: The sequential learning problem

    Michael McCloskey and Neal J Cohen. Catastrophic inter- ference in connectionist networks: The sequential learning problem. In Psychology of learning and motivation , pages 109–165. Elsevier, 1989. 2

    work page 1989

  30. [30]

    Modern hierarchical, agglomerative cluster- ing algorithms

    Daniel M ¨ullner. Modern hierarchical, agglomerative cluster- ing algorithms. arXiv preprint arXiv:1109.2378, 2011. 8

    Pith/arXiv arXiv 2011

  31. [31]

    Cats and dogs

    Omkar M Parkhi, Andrea Vedaldi, Andrew Zisserman, and CV Jawahar. Cats and dogs. In 2012 IEEE conference on computer vision and pattern recognition, pages 3498–3505. IEEE, 2012. 7

    work page 2012

  32. [32]

    Pytorch: An im- perative style, high-performance deep learning library

    Adam Paszke, Sam Gross, Francisco Massa, Adam Lerer, James Bradbury, Gregory Chanan, Trevor Killeen, Zeming Lin, Natalia Gimelshein, Luca Antiga, et al. Pytorch: An im- perative style, high-performance deep learning library. Ad- vances in neural information processing systems , 32, 2019. 5

    work page 2019

  33. [33]

    Scikit-learn: Machine learning in python

    Fabian Pedregosa, Ga ¨el Varoquaux, Alexandre Gramfort, Vincent Michel, Bertrand Thirion, Olivier Grisel, Mathieu Blondel, Peter Prettenhofer, Ron Weiss, Vincent Dubourg, et al. Scikit-learn: Machine learning in python. the Journal of machine Learning research, 12:2825–2830, 2011. 5, 8

    work page 2011

  34. [34]

    Dynamic conceptional contrastive learning for generalized category discovery

    Nan Pu, Zhun Zhong, and Nicu Sebe. Dynamic conceptional contrastive learning for generalized category discovery. In Proceedings of the IEEE/CVF conference on computer vi- sion and pattern recognition, pages 7579–7588, 2023. 6, 7

    work page 2023

  35. [35]

    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. 1, 2, 4, 5, 6, 13

    work page 2001

  36. [36]

    Learning to learn without forgetting by maximizing transfer and minimizing interference

    Matthew Riemer, Ignacio Cases, Robert Ajemian, Miao Liu, Irina Rish, Yuhai Tu, and Gerald Tesauro. Learning to learn without forgetting by maximizing transfer and minimizing interference. arXiv preprint arXiv:1810.11910, 2018. 2

    Pith/arXiv arXiv 2018

  37. [37]

    Semi-supervised self-training of object detection mod- els

    Chuck Rosenberg, Martial Hebert, and Henry Schneider- man. Semi-supervised self-training of object detection mod- els. 2005. 4

    work page 2005

  38. [38]

    Class-incremental novel class discovery

    Subhankar Roy, Mingxuan Liu, Zhun Zhong, Nicu Sebe, and Elisa Ricci. Class-incremental novel class discovery. In European Conference on Computer Vision, pages 317–333. Springer, 2022. 1, 3, 6, 7

    work page 2022

  39. [39]

    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

    work page 2015

  40. [40]

    When to Accept Automated Predictions and When to Defer to Human Judgment?

    Daniel Sikar, Artur Garcez, Tillman Weyde, Robin Bloom- field, and Kaleem Peeroo. When to accept automated predic- tions and when to defer to human judgment? arXiv preprint arXiv:2407.07821, 2024. 2, 4

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  41. [41]

    Generalized category discovery

    Sagar Vaze, Kai Han, Andrea Vedaldi, and Andrew Zisser- man. Generalized category discovery. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 7492–7501, 2022. 2

    work page 2022

  42. [42]

    A tutorial on spectral clustering.Statis- tics and computing, 17:395–416, 2007

    Ulrike V on Luxburg. A tutorial on spectral clustering.Statis- tics and computing, 17:395–416, 2007. 8

    work page 2007

  43. [43]

    Foster: Feature boosting and compression for class- incremental learning

    Fu-Yun Wang, Da-Wei Zhou, Han-Jia Ye, and De-Chuan Zhan. Foster: Feature boosting and compression for class- incremental learning. In European conference on computer vision, pages 398–414. Springer, 2022. 2, 4, 5, 6, 13

    work page 2022

  44. [44]

    Herding dynamical weights to learn

    Max Welling. Herding dynamical weights to learn. In Pro- ceedings of the 26th annual international conference on ma- chine learning, pages 1121–1128, 2009. 5

    work page 2009

  45. [45]

    Ft k-means: A high-performance k-means on gpu with fault tolerance

    Shixun Wu, Yitong Ding, Yujia Zhai, Jinyang Liu, Jiajun Huang, Zizhe Jian, Huangliang Dai, Sheng Di, Bryan M Wong, Zizhong Chen, et al. Ft k-means: A high-performance k-means on gpu with fault tolerance. In 2024 IEEE Interna- tional Conference on Cluster Computing (CLUSTER), pages 322–334. IEEE, 2024. 8, 14

    work page 2024

  46. [46]

    Large scale incre- mental learning

    Yue Wu, Yinpeng Chen, Lijuan Wang, Yuancheng Ye, Zicheng Liu, Yandong Guo, and Yun Fu. Large scale incre- mental learning. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 374–382,

    work page

  47. [47]

    Survey of clustering algo- rithms

    Rui Xu and Donald Wunsch. Survey of clustering algo- rithms. IEEE Transactions on neural networks , 16(3):645– 678, 2005. 3

    work page 2005

  48. [48]

    Class-weighted classification: Trade-offs and robust approaches

    Ziyu Xu, Chen Dan, Justin Khim, and Pradeep Raviku- mar. Class-weighted classification: Trade-offs and robust approaches. In International conference on machine learn- ing, pages 10544–10554. PMLR, 2020. 5, 13

    work page 2020

  49. [49]

    Der: Dy- namically expandable representation for class incremental learning

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

    work page

  50. [50]

    Online deep clustering for unsupervised representation learning

    Xiaohang Zhan, Jiahao Xie, Ziwei Liu, Yew-Soon Ong, and Chen Change Loy. Online deep clustering for unsupervised representation learning. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages 6688–6697, 2020. 3

    work page 2020

  51. [51]

    Flexmatch: Boosting semi-supervised learning with curricu- lum pseudo labeling

    Bowen Zhang, Yidong Wang, Wenxin Hou, Hao Wu, Jin- dong Wang, Manabu Okumura, and Takahiro Shinozaki. Flexmatch: Boosting semi-supervised learning with curricu- lum pseudo labeling. Advances in Neural Information Pro- cessing Systems, 34:18408–18419, 2021. 5

    work page 2021

  52. [52]

    mixup: Beyond empirical risk minimiza- tion

    Hongyi Zhang, Moustapha Cisse, Yann N Dauphin, and David Lopez-Paz. mixup: Beyond empirical risk minimiza- tion. arXiv preprint arXiv:1710.09412, 2017. 5, 13

    Pith/arXiv arXiv 2017

  53. [53]

    Grow and merge: A unified framework for continu- ous categories discovery

    Xinwei Zhang, Jianwen Jiang, Yutong Feng, Zhi-Fan Wu, Xibin Zhao, Hai Wan, Mingqian Tang, Rong Jin, and Yue Gao. Grow and merge: A unified framework for continu- ous categories discovery. Advances in Neural Information Processing Systems, 35:27455–27468, 2022. 1, 3, 5, 6, 7

    work page 2022

  54. [54]

    Novel visual category discov- ery with dual ranking statistics and mutual knowledge distil- lation

    Bingchen Zhao and Kai Han. Novel visual category discov- ery with dual ranking statistics and mutual knowledge distil- lation. Advances in Neural Information Processing Systems, 34:22982–22994, 2021. 2

    work page 2021

  55. [55]

    Maintaining discrimination and fairness in class incremental learning

    Bowen Zhao, Xi Xiao, Guojun Gan, Bin Zhang, and Shu- Tao Xia. Maintaining discrimination and fairness in class incremental learning. In Proceedings of the IEEE/CVF con- ference on computer vision and pattern recognition , pages 13208–13217, 2020. 2, 4, 5, 6, 13

    work page 2020

  56. [56]

    Deep class-incremental learn- ing: A survey

    Da-Wei Zhou, Qi-Wei Wang, Zhi-Hong Qi, Han-Jia Ye, De- Chuan Zhan, and Ziwei Liu. Deep class-incremental learn- ing: A survey. arXiv preprint arXiv:2302.03648, 2023. 1, 2, 3, 4, 5, 6, 13

    Pith/arXiv arXiv 2023

  57. [57]

    Class-incremental learning via dual augmentation

    Fei Zhu, Zhen Cheng, Xu-yao Zhang, and Cheng-lin Liu. Class-incremental learning via dual augmentation. In Ad- vances in Neural Information Processing Systems , pages 14306–14318. Curran Associates, Inc., 2021. 13 {OPEN} Incremental task 0 Incremental task 1 Model m0 C1 … C10 Feature extractor FC Layer C11 … C20 Training Pseudo labels generation Training ...

    work page 2021

This paper was first reviewed by deepseek-v4-flash on August 5, 2026.