Pith. sign in

REVIEW 3 major objections 5 minor 47 references

CONCLAD: COntinuous Novel CLAss Detector

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 CONCLAD detects and learns several new classes at each task using only 0.3%–1.25% of the unlabeled pool for supervision, beating earlier continual novelty detectors by large margins.

desk verdict A promising extension of incDFM to multi-class novelties with real ablations, but the central claim of separating novel classes is not actually measured and the tiny initial query has no coverage guarantee. read the letter →

arxiv 2412.10473 v1 pith:WIJJCJG2 submitted 2024-12-13 cs.LG cs.AI

Amanda Rios , Ibrahima Ndiour , Parual Datta , Omesh Tickoo , Nilesh Ahuja This is my paper · ORCID
classification cs.LGcs.AI
keywords continualnoveltydetectionmultiplenovelclassesfeaturereconstructionerrorpseudo-labelingactivelearninguncertaintyestimationout-of-distribution
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 tackles the problem of a deployed classifier that must keep learning from a stream of unlabeled data in which several brand-new classes appear together, alongside familiar classes. CONCLAD is a method that, at each task, iteratively scores every incoming sample by a ratio of old-class to novel-class feature reconstruction errors, uses that score to ask a human to label the most ambiguous samples and auto-label the most confidently novel samples, then updates its per-class models. The paper claims that with a supervision budget of only 0.3% to 1.25% of each unlabeled pool, CONCLAD separates multiple novel classes from old classes and from each other, and that it does so consistently better than existing continual novelty detectors on four datasets and two frozen feature backbones. The result matters because it removes the unrealistic oracle assumption that someone tells the model when new data is novel, while also handling more than one new class at a time, which previous continual novelty detection could not.

What carries the argument

The central object is the iterative multi-class uncertainty score S_i(u) in Equation 2, built on per-class Feature Reconstruction Error (FRE). FRE for a class m is the ℓ2 norm of the difference between a deep feature vector and its reconstruction after applying that class's PCA transform and its inverse; it measures how far a sample is from the class's feature subspace. In S_i(u), the numerator is the minimum FRE over all old classes and the denominator is the FRE of the pseudo-predicted novel class m, so the ratio orders samples from confidently novel (large numerator, small denominator) to confidently old (small numerator, large denominator). This single scalar score drives three operations: selecting ambiguous samples for the tiny active label budget, thresholding confidently novel samples for pseudo-labeling, and updating per-class PCA transforms and the pseudo-labeling head at each inner-loop iteration.

What would settle it

For a dataset with several novel classes per task, train each class's PCA subspace using only the samples CONCLAD would pseudo-label from a 0.3% budget, then measure the mean FRE of held-out samples from that same class versus held-out samples from other classes; the central ordering claim is falsified if, for any true novel class, held-out samples of that class do not have the lowest mean FRE under that class's own subspace.

Watch

Extended reading notes

Core claim

The paper's central claim is that generalized continual novel class detection—multiple new classes arriving together at each task, mixed with unseen old-class samples and no task oracle—becomes tractable if uncertainty is measured iteratively by a ratio of feature reconstruction errors. Starting from per-class PCA models of the old classes, CONCLAD first labels a few ambiguous samples, builds a rough PCA subspace and a pseudo-labeling head for each detected novel class, then recomputes the score S_i(u) = min_{j in old} FRE_j(u) / FRE_m(u) for the pseudo-predicted novel class m. High scores mark confidently novel samples for pseudo-labeling; low scores mark old-class samples; mid-range scores are actively queried. Over inner-loop iterations the per-class subspaces and the pseudo-labeler improve, and the final models are folded into the detector so that the new classes become old classes in the next task. Averaged AUROC numbers reported for ImageNet-21K-OOD, Eurosat, iNaturalist-Plants-20, and Cifar100 place CONCLAD above incDFM, DFM, ER, and PseudoER across almost all settings, with ablations indicating that the iterative loop, the pseudo-labeling, and the ambiguous-sample querying all contribute to the gain.

Load-bearing premise

The load-bearing premise is that a per-class PCA transform estimated from a tiny handful of labeled and pseudo-labeled samples (as few as 0.3% of the pool) faithfully represents the class's true feature distribution, so that the ratio in Equation 2 orders confident novel from confident old; if a pseudo-labeled sample is misassigned, its polluted subspace can systematically misroute later samples.

Editorial extensions

If this is right

  • With a supervision budget as small as 0.3% of the unlabeled pool, several new classes can be discovered, separated, and learned within one task instead of just one new class per task.
  • Because each novel class receives its own PCA transform, updates are incremental: classes discovered at task t are added without disturbing older transforms and become old classes at task t+1.
  • Querying ambiguous samples near the detection threshold contributes to the gains; the ablations show it beats querying the most-confident novel samples (Sup-Top) by 2.4% average AUROC.
  • The iterative inner loop is essential: removing it costs an average 11.2% AUROC, and removing pseudo-labeling costs 16.8%, indicating both serve to limit continual error propagation.

Reading between the lines

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

  • Because the score design is agnostic to the underlying uncertainty metric, one could re-instantiate Equation 2 with Mahalanobis distances or likelihood ratios instead of FRE and test whether the ordering property persists with fewer labeled samples.
  • The method's reliance on a frozen, high-quality feature extractor suggests the reported gains may depend on representation quality; a natural stress test is to run CONCLAD with a smaller or fine-tuned backbone and see how much of the AUROC gap remains.
  • The first inner iteration has no novel-class models, so the initial ambiguity sampling is inherently heuristic; an adaptive querying rule that expands candidate coverage as novel classes are discovered might improve early-task accuracy on very large pools.
  • Since the authors list old-class distribution shift as future work, a testable extension is to augment the FRE ratio with a drift-aware term so the detector can distinguish a genuinely new class from an old class whose appearance has shifted.
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 CONCLAD, a continual novelty detection method that combines per-class PCA feature reconstruction error (FRE), an iterative uncertainty scoring function, active supervision from a tiny labeling budget, and pseudo-labeling of confident novel samples. The method is evaluated on four datasets with frozen ResNet50 and ViT backbones, and compared against continual novelty detection baselines and ablations. The central claim is that CONCLAD continuously separates old-class samples from multiple novel classes with high AUROC while using only 0.3% to 1.25% of each task's unlabeled pool for supervision.

Significance. If the claimed results hold, CONCLAD addresses a realistic and under-explored problem: continual novel class detection when several new classes can appear at once and no task oracle is available. The use of frozen feature extractors, the independent held-out test evaluation, the inclusion of official baseline implementations, and the ablations isolating the iterative and pseudo-labeling components are strengths. The main significance gap is that the paper's distinctive claim of discriminating among multiple novel classes is not directly measured: all reported metrics are binary old-versus-new AUROC, which would also be high for a method that collapses all novel classes into one. The lack of uncertainty quantification and missing reproducibility details further limit the strength of the numerical claims.

major comments (3)
  1. [Section 3.2 and Fig. 2 (left)] The central claim that CONCLAD discriminates between different novel classes themselves is not supported by any reported metric. AUROC is a binary old-versus-new measure; it does not test whether samples from two different novel classes are routed to distinct per-class PCA models or pseudo-labels. Since Eq. (2) depends on the pseudo-labeler assigning each sample to the correct novel class, the paper needs a direct inter-novel-class metric, such as per-class recall, clustering purity, or a confusion matrix among novel classes. Without this, the reported AUROC advantages could be obtained by a method that treats all novel samples as one class, which is exactly the failure mode the paper attributes to incDFM.
  2. [Section 2.2.2, first inner iteration] The initial supervision query samples uniformly among instances with S0(u) above the mean S0(u), and these labels initialize both the pseudo-labeler and the per-novel-class PCA transforms. With a 0.3% budget and several novel classes per task, a novel class can easily be absent from this high-S0 tail. The paper acknowledges that not all novel classes may be found at i=0, but no recovery mechanism is described: at subsequent iterations the pseudo-labeler has output nodes only for already-detected classes, and Eq. (2) routes a sample of a missed class through the nearest initialized class, making its score low and preventing detection. The authors should either provide a coverage guarantee or a concrete mechanism for later iterations to discover skipped classes, and report per-class discovery rates.
  3. [Section 3.1 and Appendix 4.2.1] The manuscript reports no error bars, no number of independent runs, and no standard deviations, even though several reported margins are small (e.g., Plants R50: CONCLAD 73.6 vs. incDFM 68.7; Plants ViT: CONCLAD 58.8 vs. incDFM 58.2). In addition, the per-class PCA subspace dimension is never stated, although it is a critical parameter of the FRE scoring function in Eq. (3). The authors should report means and variances over multiple runs, and state the PCA dimension and the hyperparameter selection procedure, to substantiate the claim of consistent and large margins.
minor comments (5)
  1. [Section 3.2] The center panel of Fig. 1 is referenced as 'figure X', and Fig. 1 itself appears to be missing from the manuscript; this placeholder must be fixed before the paper can be properly reviewed.
  2. [Abstract, Section 2.2, Section 3.1] The supervision budget is stated inconsistently as 0.3% to 1.25% in the abstract, 0.3% to 2.5% in Section 2.2, and as varying from 0.32% to 5% in Section 3.1; please unify these numbers and clarify the budget used for each experiment.
  3. [Section 3.1 and Section 2.2.2] The pseudo-labeling head is denoted N_pl in Section 2.2.2 but N_ps and Acl_s in Section 3.1; please use consistent notation throughout.
  4. [Section 4.1.1] The inner-loop stopping threshold Tinner and the ambiguity formulation depend on an in-distribution validation set, but the paper does not state the size of this validation set or how its 0.1% subset is selected; please provide these details.
  5. [Appendix 4.3] The caption for Fig. 3 lists panels D/E for Plants and Cifar100, but the corresponding dataset selection and task design are not described as precisely as for the main datasets; please add the same setup details as in Section 3.1.

Circularity Check

0 steps flagged · score 2.0 of 10

No significant circularity: the central AUROC claim rests on held-out evaluation; mild self-citation of FRE and incDFM is not load-bearing.

full rationale

The derivation chain is self-contained in the sense required by the circularity check. CONCLAD's uncertainty score S_i(u) in Eq. (2) is a stated scoring rule: a ratio of the FRE distance to old classes over the FRE distance to the pseudo-labeled novel class. The paper's headline claim is that this rule, combined with iterative active querying and pseudo-labeling, achieves high AUROC on held-out test sets. That evaluation uses ground-truth old/new labels and therefore does not reduce to the definition of S_i. FRE is imported from prior work [14] by overlapping authors, and incDFM [11] is a self-authored baseline, but FRE is an externally published, falsifiable subspace-reconstruction method and incDFM is used as a comparison baseline, not as a uniqueness theorem or as the justification for the central result. No fitted parameter is renamed as a prediction: the tiny supervision budget is used to fit per-class PCA transforms and a pseudo-labeler, and the reported AUROC is measured on an independent test set with the same old-to-new ratio. The manuscript does carry real limitations—only binary AUROC is reported even though the abstract claims discrimination between novel classes, Fig. 1 is missing, and the text refers to a placeholder 'figure X'—but these are evaluation-coverage and presentation issues, not circular derivation. Accordingly, no circular step is identified; the score of 2 reflects only the mild authorship-overlap burden in the uncertainty core and the closest baseline.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The central algorithm adds no new physical entities. Its performance depends on the calibration of FRE/PCA subspaces on frozen features, on the hand-set thresholds alpha and Tinner, and on the supervised budget. These are assumptions, not derived results.

free parameters (5)
  • alpha (pseudo-label selection fraction) = 0.20
    Selects the top 20% of samples predicted novel for automatic pseudo-labeling; fixed across experiments, no sensitivity analysis shown (Appendix 4.1.1).
  • Tinner stopping threshold = 2 standard deviations above in-distribution validation mean
    Determines which samples are considered novel and defines the ambiguity region; asserted insensitive without supporting plot (Appendix 4.1.1).
  • supervision labeling budget = 0.625% (Eurosat, Im21K-OOD) and 1.25% (Plants, CIFAR-100)
    Experimental variable, but final performance and the claim of tiny budget depend on this choice (Section 3.1, Fig. 1 center).
  • per-class PCA subspace dimension = not specified
    FRE reconstruction error depends on how many principal components are kept; not reported, so the main uncertainty metric is partially underspecified.
  • old/new class mixing ratio in task pools = 2:1 old to new classes
    The composition of each unlabeled pool is chosen by the authors and directly controls the AUROC evaluation (Section 3.1).
assumptions (4)
  • domain assumption Frozen pretrained features (SwAV/DINO) are task-nonspecific and sufficiently informative for continual novelty detection without updating the backbone.
    Invoked throughout Section 3.1 and Appendix 4.2.1; if false, per-class PCA on frozen embeddings cannot represent novel classes.
  • domain assumption FRE reconstructability via per-class PCA is a reliable class-conditional density proxy, so low reconstruction error implies membership in that class.
    Section 2.2.1 and Appendix 4.1.3; this is the core uncertainty estimate on which Eq. (2) rests.
  • domain assumption The in-distribution validation set built from held-out previous-task predictions reflects the old-class feature distribution for thresholding.
    Appendix 4.1.1 defines Tinner using a 0.1% held-out validation set; errors in this estimate shift all decisions.
  • ad hoc to paper Initial uniform sampling among highest S0 scores will label at least one sample from each, or enough, novel classes to initialize per-class PCA transforms.
    Section 2.2.2, i=0 step; missing a novel class at initialization may prevent it from being discovered later, as the authors only note that some classes may be found in subsequent iterations.

how reviews work

0 comments
Cite this review

Pith. "Pith review of CONCLAD: COntinuous Novel CLAss Detector." pith.science (2026). https://pith.science/paper/WIJJCJG2

@misc{pith2026241210473,
  author       = {Pith},
  title        = {Pith review of: CONCLAD: COntinuous Novel CLAss Detector},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/WIJJCJG2}},
  note         = {Machine review of arXiv:2412.10473}
}
read the original abstract

In the field of continual learning, relying on so-called oracles for novelty detection is commonplace albeit unrealistic. This paper introduces CONCLAD ("COntinuous Novel CLAss Detector"), a comprehensive solution to the under-explored problem of continual novel class detection in post-deployment data. At each new task, our approach employs an iterative uncertainty estimation algorithm to differentiate between known and novel class(es) samples, and to further discriminate between the different novel classes themselves. Samples predicted to be from a novel class with high-confidence are automatically pseudo-labeled and used to update our model. Simultaneously, a tiny supervision budget is used to iteratively query ambiguous novel class predictions, which are also used during update. Evaluation across multiple datasets, ablations and experimental settings demonstrate our method's effectiveness at separating novel and old class samples continuously. We will release our code upon acceptance.

Figures

Figures reproduced from arXiv: 2412.10473 by the authors.

Figure 1
Figure 1. (Left A.1,B.1) Continual Novelty Detection performance measured by AUROC at each task. The number of novel classes introduced per task is in parenthesis. Overall, CONCLAD (green) significantly over-performs baselines; (Center A.2,B.2) Results varying the supervision budget; (Right A.3,B.3) Results varying Novel Class Increment per task. For (left,right) Supervision budget is 0.625% for CONCLAD, ER, PseudoER. Equival… view at source ↗
Figure 2
Figure 2. (Left) Continual Novelty Detection measured by AUROC; (Right) Ablations of CONCLAD. [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Results for Plants and Cifar100; (Left D.1,E.1) Continual Novelty Detection performance measured by AUROC at each task. The number of novel classes introduced per task is in paren￾thesis.(Center D.2,E.2) Results varying the supervision budget; (Right D.3,E.3) Results varying Novel Class Increment per task. For (left,right) Supervision budget is 1.25% for CONCLAD, ER, PseudoER. Overall, CONCLAD (green) significantly … view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

47 extracted references · 37 canonical work pages

  1. [1]

    Overcoming catastrophic forgetting in neural networks

    James Kirkpatrick et al. “Overcoming catastrophic forgetting in neural networks”. In: Proceed- ings of the national academy of sciences 114.13 (2017), pp. 3521–3526

    work page 2017

  2. [2]

    Beneficial Perturbation Network for designing general adaptive artificial intelligence systems

    Shixian Wen et al. “Beneficial Perturbation Network for designing general adaptive artificial intelligence systems”. In: IEEE Transactions on Neural Networks and Learning Systems (2021)

    work page 2021

  3. [3]

    Superposition of many models into one

    Brian Cheung et al. “Superposition of many models into one”. In: Advances in neural informa- tion processing systems 32 (2019)

    work page 2019

  4. [4]

    icarl: Incremental classifier and representation learning

    Sylvestre-Alvise Rebuffi et al. “icarl: Incremental classifier and representation learning”. In: Proceedings of the IEEE conference on Computer Vision and Pattern Recognition . 2017, pp. 2001–2010

    work page 2017

  5. [5]

    Lifelong Learning Without a Task Oracle

    Amanda Rios and Laurent Itti. “Lifelong Learning Without a Task Oracle”. In: 2020 IEEE 32nd International Conference on Tools with Artificial Intelligence (ICTAI) . IEEE. 2020, pp. 255–263

    work page 2020

  6. [6]

    Rethinking experience replay: a bag of tricks for continual learning

    Pietro Buzzega et al. “Rethinking experience replay: a bag of tricks for continual learning”. In: 2020 25th International Conference on Pattern Recognition (ICPR). IEEE. 2021, pp. 2180– 2187

    work page 2020

  7. [7]

    Continual lifelong learning with neural networks: A review

    German I Parisi et al. “Continual lifelong learning with neural networks: A review”. In:Neural Networks 113 (2019), pp. 54–71

    work page 2019

  8. [8]

    A soft nearest-neighbor framework for continual semi-supervised learn- ing

    Zhiqi Kang et al. “A soft nearest-neighbor framework for continual semi-supervised learn- ing”. In: Proceedings of the IEEE/CVF International Conference on Computer Vision. 2023, pp. 11868–11877

    work page 2023

Show all 47 references
  1. [9]

    Continual semi-supervised learning through contrastive interpolation consistency

    Matteo Boschini et al. “Continual semi-supervised learning through contrastive interpolation consistency”. In: Pattern Recognition Letters162 (Oct. 2022), pp. 9–14. ISSN : 0167-8655. DOI: 10.1016/j.patrec.2022.08.006 . URL: http://dx.doi.org/10.1016/j.patrec. 2022.08.006

    2022 doi

  2. [10]

    Beyond Supervised Continual Learning: a Review

    Benedikt Bagus, Alexander Gepperth, and Timothée Lesort. Beyond Supervised Continual Learning: a Review. 2022. arXiv: 2208.14307 [cs.LG]

    2022 arXiv

  3. [11]

    incdfm: Incremental deep feature modeling for continual novelty detec- tion

    Amanda Rios et al. “incdfm: Incremental deep feature modeling for continual novelty detec- tion”. In: European Conference on Computer Vision. Springer. 2022, pp. 588–604

    2022

  4. [12]

    A baseline for detecting misclassified and out-of- distribution examples in neural networks

    Dan Hendrycks and Kevin Gimpel. “A baseline for detecting misclassified and out-of- distribution examples in neural networks”. In: (2017)

    2017

  5. [13]

    Enhancing the reliability of out-of-distribution image detection in neural networks

    Shiyu Liang, Yixuan Li, and R Srikant. “Enhancing the reliability of out-of-distribution image detection in neural networks”. In: (2018)

    2018

  6. [14]

    Out-Of-Distribution Detection With Subspace Techniques And Probabilistic Modeling Of Features

    Ibrahima Ndiour, Nilesh A Ahuja, and Omesh Tickoo. “Out-Of-Distribution Detection With Subspace Techniques And Probabilistic Modeling Of Features”. In: arXiv preprint arXiv:2012.04250 (2020)

    2020 arXiv

  7. [15]

    A simple unified framework for detecting out-of-distribution samples and adversarial attacks

    Kimin Lee et al. “A simple unified framework for detecting out-of-distribution samples and adversarial attacks”. In: Advances in Neural Information Processing Systems. 2018, pp. 7167– 7177

    2018

  8. [16]

    ViM: Out-Of-Distribution with Virtual-logit Matching

    Haoqi Wang et al. “ViM: Out-Of-Distribution with Virtual-logit Matching”. In:Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2022

    2022

  9. [17]

    Likelihood ratios for out-of-distribution detection

    Jie Ren et al. “Likelihood ratios for out-of-distribution detection”. In: Advances in Neural Information Processing Systems. 2019, pp. 14707–14718

    2019

  10. [18]

    Deep clustering: A comprehensive survey

    Yazhou Ren et al. “Deep clustering: A comprehensive survey”. In: IEEE Transactions on Neural Networks and Learning Systems (2024)

    2024

  11. [19]

    How to measure uncertainty in uncertainty sampling for active learning

    Vu-Linh Nguyen, Mohammad Hossein Shaker, and Eyke Hüllermeier. “How to measure uncertainty in uncertainty sampling for active learning”. In: Machine Learning 111.1 (2022), pp. 89–122. 5

    2022

  12. [20]

    Deep bayesian active learning with image data

    Yarin Gal, Riashat Islam, and Zoubin Ghahramani. “Deep bayesian active learning with image data”. In: International Conference on Machine Learning. PMLR. 2017, pp. 1183–1192

    2017

  13. [21]

    Learning loss for active learning

    Donggeun Yoo and In So Kweon. “Learning loss for active learning”. In:Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2019, pp. 93–102

    2019

  14. [22]

    ImageNet-21K Pretraining for the Masses

    Tal Ridnik et al. ImageNet-21K Pretraining for the Masses . 2021. arXiv: 2104 . 10972 [cs.CV]

    2021

  15. [23]

    EuroSAT: A Novel Dataset and Deep Learning Benchmark for Land Use and Land Cover Classification

    Patrick Helber et al. EuroSAT: A Novel Dataset and Deep Learning Benchmark for Land Use and Land Cover Classification. 2019. arXiv: 1709.00029 [cs.CV]

    2019 arXiv

  16. [24]

    The iNaturalist Species Classification and Detection Dataset

    Grant Van Horn et al. The iNaturalist Species Classification and Detection Dataset . 2018. arXiv: 1707.06642 [cs.CV]

    2018 arXiv

  17. [25]

    Learning Multiple Layers of Features from Tiny Images

    Alex Krizhevsky. “Learning Multiple Layers of Features from Tiny Images”. In: University of Toronto(May 2012)

    2012

  18. [26]

    Subspace Modeling for Fast Out- Of-Distribution and Anomaly Detection

    Ibrahima J Ndiour, Nilesh A Ahuja, and Omesh Tickoo. “Subspace Modeling for Fast Out- Of-Distribution and Anomaly Detection”. In: 2022 IEEE International Conference on Image Processing (ICIP). IEEE. 2022, pp. 3041–3045

    2022

  19. [27]

    Experience replay for continual learning

    David Rolnick et al. “Experience replay for continual learning”. In: Advances in Neural Information Processing Systems 32 (2019)

    2019

  20. [28]

    Continual novelty detection

    Rahaf Aljundi et al. “Continual novelty detection”. In: Conference on Lifelong Learning Agents. PMLR. 2022, pp. 1004–1025

    2022

  21. [29]

    Pseudo-Label : The Simple and Efficient Semi-Supervised Learning Method for Deep Neural Networks

    Dong-Hyun Lee. “Pseudo-Label : The Simple and Efficient Semi-Supervised Learning Method for Deep Neural Networks”. In:ICML 2013 Workshop : Challenges in Representation Learning (WREPL) (July 2013)

    2013

  22. [30]

    Deep residual learning for image recognition

    Kaiming He et al. “Deep residual learning for image recognition”. In: Proceedings of the IEEE conference on computer vision and pattern recognition. 2016, pp. 770–778

    2016

  23. [31]

    Unsupervised learning of visual features by contrasting cluster as- signments

    Mathilde Caron et al. “Unsupervised learning of visual features by contrasting cluster as- signments”. In: Advances in Neural Information Processing Systems 33 (2020), pp. 9912– 9924

    2020

  24. [32]

    An image is worth 16x16 words: Transformers for image recognition at scale

    Dosovitskiy Alexey. “An image is worth 16x16 words: Transformers for image recognition at scale”. In: arXiv preprint arXiv: 2010.11929 (2020)

    2020 arXiv

  25. [33]

    Emerging properties in self-supervised vision transformers

    Mathilde Caron et al. “Emerging properties in self-supervised vision transformers”. In: Pro- ceedings of the IEEE/CVF international conference on computer vision . 2021, pp. 9650– 9660

    2021

  26. [34]

    Generalized odin: Detecting out-of-distribution image without learning from out-of-distribution data

    Yen-Chang Hsu et al. “Generalized odin: Detecting out-of-distribution image without learning from out-of-distribution data”. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2020, pp. 10951–10960

    2020

  27. [35]

    Head2toe: Utilizing intermediate representations for better transfer learning

    Utku Evci et al. “Head2toe: Utilizing intermediate representations for better transfer learning”. In: International Conference on Machine Learning. PMLR. 2022, pp. 6009–6033

    2022

  28. [36]

    The dynamics of perceptual learning: an incremental reweighting model

    Alexander A Petrov, Barbara Anne Dosher, and Zhong-Lin Lu. “The dynamics of perceptual learning: an incremental reweighting model.” In: Psychological review 112.4 (2005), p. 715

    2005

  29. [37]

    A baseline for few-shot image classification

    Guneet S Dhillon et al. “A baseline for few-shot image classification”. In: arXiv preprint arXiv:1909.02729 (2019)

    2019 arXiv

  30. [38]

    Adam: A method for stochastic optimization

    Diederik P Kingma and Jimmy Ba. “Adam: A method for stochastic optimization”. In: arXiv preprint arXiv:1412.6980 (2014)

    2014 arXiv

  31. [39]

    Closed-loop memory GAN for continual learning

    Amanda Rios and Laurent Itti. “Closed-loop memory GAN for continual learning”. In: arXiv preprint arXiv:1811.01146 (2018)

    2018 arXiv

  32. [40]

    ImageNet Large Scale Visual Recognition Challenge

    Olga Russakovsky et al. “ImageNet Large Scale Visual Recognition Challenge”. English (US). In: International Journal of Computer Vision 115.3 (Dec. 2015). Publisher Copyright: © 2015, Springer Science+Business Media New York., pp. 211–252. ISSN : 0920-5691. DOI: 10.1007/s11263...

    2015 doi

  33. [41]

    MOS: Towards Scaling Out-of-distribution Detection for Large Semantic Space

    Rui Huang and Yixuan Li. “MOS: Towards Scaling Out-of-distribution Detection for Large Semantic Space”. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2021

    2021

  34. [42]

    Delving into Out-of-Distribution Detection with Vision-Language Represen- tations

    Yifei Ming et al. “Delving into Out-of-Distribution Detection with Vision-Language Represen- tations”. In: Advances in Neural Information Processing Systems. 2022

    2022

  35. [43]

    A survey of deep active learning

    Pengzhen Ren et al. “A survey of deep active learning”. In: ACM computing surveys (CSUR) 54.9 (2021), pp. 1–40. 6 4 Appendix 4.1 Methodology Details 4.1.1 Thresholds for Stopping the inner-loop The inner-loop is guided by two simple thresholds: (1) Threshold Tinner "roughly" e...

    2021

  36. [44]

    We use a random set of 500 samples from each of the 50 classes

    Imagenet21K-OOD (Im21K-OOD) [22]: We curated a subset of Imagenet21K containing the top-most populous 50 classes and that do not overlap with the classes present in Imagenet1K. We use a random set of 500 samples from each of the 50 classes. Because Imagenet21K is a superset of...

  37. [45]

    A super-set (larger) version of this subset was originally proposed by [41] and has since been frequently used as test OOD dataset with respect to Imagenet1K [42]

    iNaturalist-Plants-20 (Plants) [24]: is a curated subset containing images from 20 OOD plant species, sourced from the iNaturalist project [24]. A super-set (larger) version of this subset was originally proposed by [41] and has since been frequently used as test OOD dataset w...

  38. [46]

    Eurosat [23]: An RGB dataset of 10 classes and 27K images of Sentinel-2 satellite images, which is also orthogonal to Imagenet1K

  39. [47]

    This totals 20 labels (super) and 50K images

    Cifar100-Superclasses (Cifar100) [25]: We use the super-label granularity of Cifar100 dataset. This totals 20 labels (super) and 50K images. While Cifar100 is not orthogonal to Imagenet1K, we decided to showcase its results since it is a widespread dataset in CL. 4.2.3 Baselin...

Pith tools

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