Pith. sign in

REVIEW 4 major objections 6 minor 78 references

Open set label noise learning with robust sample selection and margin-guided module

T0 review · 4 major / 6 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read This paper claims that combining a two-stage robust sample selection with margin-based separation of out-of-distribution noise yields higher test accuracy than prior label-noise learning methods on both synthetic and real-world image…

desk verdict A plausible incremental LNL recipe that overclaims its core innovation: the margin-guided ID/OOD split is never directly measured, and PNP beats it on two key settings. read the letter →

arxiv 2501.04269 v1 pith:2AUFXKIY submitted 2025-01-08 cs.CV

Yuandi Zhao , Qianxi Xia , Yang Sun , Zhijie Wen , Liyan Ma , Shihui Ying This is my paper · ORCID
classification cs.CV
keywords opensetlabelnoiselearningrobustsampleselectionmargin-guidedmoduleout-of-distributiondetectionsemi-supervisedrelabelingnoisyimageclassificationCIFAR-100N
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 addresses a realistic failure mode in image classification: training labels contain both closed-set errors, where the true class is in the known label list, and open-set errors, where the image belongs to an unknown class. The authors propose RSS-MGM, which first selects a larger set of clean samples by taking the union of small-loss selection and high-confidence selection, then divides the remaining noisy samples by margin rules. Open-set noise is discarded, while confident in-distribution noise is re-labeled through sharpened pseudo-labels and trained with a consistency loss. The claim is that this multi-level reuse of noisy data outperforms existing label-noise learning methods on CIFAR100N-C, CIFAR80N-O, WebFG-496, and Food101N, and separates open-set from closed-set noise more accurately.

What carries the argument

The load-bearing mechanism is the two-module partition of the training set. The robust sample selection module computes $d_i = D_{JS}(p_i \| y_i)$ and a confidence score $s_i$, taking $\mathcal{D}_{\mathrm{clean}} = \{x_i : 1-d_i > \tau_s\} \cup \{x_i : s_i > \tau_h\}$, which widens the clean set beyond small-loss filtering. The margin-guided module then works on the noise set, using the weak/strong augmentation disagreement margin $M_v = \mathbb{I}[\arg\max_c p^w \neq \arg\max_c p^s]$ to flag out-of-distribution samples, and the average top-one-minus-top-two logit margin $M_p = \frac{1}{2}[(z_y(v^w)-\max_{k\neq y} z_k(v^w)) + (z_y(v^s)-\max_{k\neq y} z_k(v^s))]$ to select confident in-distribution noise for pseudo-label relabeling. These two margins carry the argument: they decide which samples are discarded, which are re-labeled, and which are trained as clean.

What would settle it

On the CIFAR80N-O benchmark, record which known open-set samples are placed in $\mathcal{D}_{\mathrm{OOD}}$, $\mathcal{D}_{\mathrm{high}}$, and $\mathcal{D}_{\mathrm{rest}}$; if a substantial fraction of true OOD samples are kept as in-distribution and re-labeled, or a substantial fraction of clean or closed-set samples are discarded, the margin-guided separation is not working as claimed and the accuracy gains must be attributed to other components.

Watch

Extended reading notes

Core claim

The central claim is that label noise should not be treated as all-or-nothing: instead of deleting everything that fails a clean-sample test, a network can profitably re-use part of the noise. RSS-MGM keeps the union of samples with low Jensen-Shannon divergence between prediction and label and samples whose predicted probability on the annotated class is high, giving a larger clean set than small-loss selection alone. The remaining noise set is split by two margin functions: a sample is treated as out-of-distribution if the argmax of the weakly augmented and strongly augmented predictions disagree, and as in-distribution otherwise; among in-distribution noise, samples with a large margin between the top and second logit are re-labeled via temperature-sharpened pseudo-labels and trained with a consistency-regularized loss, while the rest are dropped. On the reported benchmarks this pipeline beats or matches the strongest published methods, with the largest absolute gains at high noise rates.

Load-bearing premise

The pipeline rests on the behavioral assumption that open-set noisy images produce near-uniform prediction probabilities whose argmax flips between weak and strong augmentation, while in-distribution images produce confident, stable one-hot predictions.

Editorial extensions

If this is right

  • At 80% symmetric noise on CIFAR100N-C, RSS-MGM reports 41.88% test accuracy versus 31.32% for PNP, so the high-confidence selection and relabeling are most valuable when few clean labels remain.
  • On CIFAR80N-O, the reported accuracies of 67.32%, 61.59%, and 39.86% at 20%, 50%, and 80% noise are higher than all compared methods, indicating the OOD-discarding rule helps in open-set settings even at high noise.
  • On Web-Aircraft, Web-Bird, and Web-Car, RSS-MGM reaches 85.82%, 82.34%, and 90.15%, beating the strongest published baseline PNP on all three real-world web datasets.
  • Ablations show that removing both RSS and MGM drops accuracy from 67.32% to 62.47% on CIFAR80N-O with 20% symmetric noise, and removing the margin functions $\mathcal{M}_v$ and $\mathcal{M}_p$ separately drops accuracy to 64.75% and 65.38%.

Reading between the lines

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

  • The paper does not report precision or recall for the OOD-versus-ID partition; a direct test would be to count true open-set samples that survive into the relabeled set, and if that fraction is high, the accuracy gains may come more from loss robustness than from accurate margin-guided separation.
  • The margin rule assumes OOD predictions are near-uniform, so the method's advantage should shrink when open-set classes resemble in-distribution classes and produce confident but wrong predictions.
  • The same triage idea could be applied to other data-quality problems such as webly supervised fine-grained recognition, where margin disagreement is used to identify mislabeled images before pseudo-labeling; this is an extension rather than part of the paper's experiments.
  • A testable consequence of the paper's claim is that replacing the weak/strong disagreement rule with any uncertainty-based OOD detector should yield similar or better gains, and the gap should be measurable on WebFG-496.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 6 minor

Summary. The paper proposes RSS-MGM, a training scheme for image classification under label noise that mixes both closed-set and open-set noise. The method first builds a larger clean set by unioning a small-loss (JS-divergence) selection with a high-confidence selection, then applies a margin-guided module that splits the remaining noisy samples into OOD (discarded) and ID (re-labeled) sets using two augmented views and a margin ranking function. The final loss combines cross-entropy on clean samples, an improved cross-entropy on high-confidence ID samples, and a consistency regularizer. The authors evaluate on CIFAR100N-C, CIFAR80N-O, WebFG-496, and Food101N and report accuracy gains over several prior methods, with ablations on CIFAR80N-O.

Significance. If the claimed ID/OOD separation were convincingly demonstrated, RSS-MGM would be a useful contribution to the open-set label-noise literature, because it attempts to exploit discarded noisy samples rather than simply dropping them. The robust sample-selection idea of combining small-loss and high-confidence criteria is plausible and easy to reproduce, and the paper includes comparisons on both synthetic and real-world benchmarks. However, the paper's headline claim that RSS-MGM 'more accurately divides open set label noise samples and closed set ones' is never directly measured, and on two of the ten synthetic-noise settings the method is outperformed by PNP. The contribution is therefore currently conditional on additional evidence; the overall significance is moderate rather than definitive.

major comments (4)
  1. [§4.2, Tables 1 and 2] In two of the ten reported synthetic-noise configurations the method does not outperform PNP: on CIFAR100N-C Asymmetric-40% RSS-MGM gives 58.36% versus PNP's 60.25%, and on CIFAR80N-O Asymmetric-40% RSS-MGM gives 59.80% versus PNP's 61.23%. The surrounding text nevertheless states that 'RSS-MGM outperforms the prior PNP approach' and claims general superiority over state-of-the-art methods. Please qualify the claim to report the configurations where the method wins and loses, and add error bars or significance statements for the real-world tables, which currently have none.
  2. [§3.2, Eqs. (11)-(13), and §4.4] The central claim that RSS-MGM 'more accurately divides open set label noise samples and closed set ones' is not supported by any direct measurement. The OOD decision rule in Eq. (11) assumes that OOD samples produce near-uniform predictions whose argmax flips between weak and strong augmentations, while ID samples produce stable one-hot predictions, but no experiment in Section 4 reports precision, recall, or F1 for the ID/OOD split, even though CIFAR80N-O provides ground-truth OOD membership by construction. This is particularly important because OOD samples in CIFAR80N-O are ordinary CIFAR-100 images from 20 held-out classes, and a classifier trained on 80 classes may map them confidently and consistently to a nearby known class, making M_v=0 and causing them to be retained as ID; the ablations in Table 6 only report final accuracy, which can change through the selection or consistency-loss channels without validating the separation behavior.
  3. [§3.4, Eq. (24), versus §3.3, Eqs. (21)-(22)] The re-labeling mechanism is not consistently defined. Section 3.3 states that samples in D_high receive pseudo-labels computed by sharpening (Eq. (21)) and averaging the two views (Eq. (22)), but the loss in Eq. (24) is written as a cross-entropy against y_i^c, which the text identifies as 'the ground-truth label of sample x_i'. If the original noisy label is used, the re-labeling step is not actually implemented; if a pseudo-label is intended, the notation should be changed to q_i^c and the sum should range over D_high, not D_id as written. This ambiguity is load-bearing because the claimed benefit of 'correcting ID samples' depends on which label the loss uses.
  4. [§4.1, thresholds in Eqs. (6), (8), and (18)] The selection thresholds tau_s=0.75, tau_h=0.9, and tau_p=0.9 are fixed values with no sensitivity analysis or selection criterion. These thresholds directly determine the sizes of the clean, ID, and OOD sets, and the method's behavior could change substantially with their values. Please report the performance as a function of each threshold, or justify the chosen values with a principled rule, since otherwise the reported gains may be threshold-tuned rather than intrinsic to the method.
minor comments (6)
  1. [Abstract and §4] The dataset name is inconsistent: the abstract and several places in the text say 'WebFG-469', while Section 4 and the Data Availability statement say 'WebFG-496'; please correct this.
  2. [Algorithm 1 caption] The caption reads 'RSS-MGS Algorithm'; this should be 'RSS-MGM'.
  3. [§3.2, Eq. (12)] The set-builder notation in Eq. (12), D_OOD = {x_i | M_i^v · D_n, x_i in D_n}, is malformed; it should be written as D_OOD = {x_i in D_n | M_i^v = 1}.
  4. [§4.3] The heading 'Food01N' is a typo for 'Food101N'.
  5. [§4.1, experimental settings] The sentence 'The selection coefficients λ1 and λ2 are set to 0.05 each are both set to 0.05' is duplicated and should be rewritten.
  6. [§4.2, citation] The sentence 'The results of existing approaches are sourced from Jo-SRC (Wei et al., 2020)' appears to cite the wrong paper: Jo-SRC is by Yao et al. (2021), while Wei et al. (2020) is JoCoR; please check this citation.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the paper's accuracy claims are empirical benchmark comparisons, and no equation reduces a prediction to a fitted input or to a self-citation chain.

full rationale

RSS-MGM's central claim is that the proposed sample-selection and margin-guided pipeline achieves higher test accuracy than prior methods on CIFAR100N-C, CIFAR80N-O, WebFG-496, and Food101N. This is an empirical claim evaluated against held-out test sets, not a derivation from the method's own definitions. The robust sample selection module (Eqs. 3-10) defines a clean set as a union of a JS-divergence-selected set and a confidence-thresholded set; the statement that this yields 'more clean samples' is a set-size tautology, but the reported accuracy is not constructed from that definition. The margin-guided module (Eqs. 11-18) operationally defines OOD samples as those whose argmax flips between weak and strong augmentations; this is an assumption about OOD behavior, and the paper does not directly measure ID/OOD separation precision or recall, but the absence of that measurement is an evidentiary gap, not a circular reduction. The only self-citation (Wen et al., 2023, JSMix) appears in a related-work list and is not used to justify the correctness of the present method; the JS-divergence criterion is stated and motivated in the paper itself. Thresholds tau_s, tau_h, and tau_p are hyperparameters, and no result is presented as a prediction of a quantity that was itself used to fit them. Consequently, the paper does not exhibit any step where an output is equivalent to an input by construction or where a fitted parameter is renamed as a prediction.

Assumptions & free parameters 7 free parameters · 5 assumptions · 0 invented entities

The empirical claims rest on hand-set thresholds and on unverified behavioral assumptions about DNN predictions; no new entities are introduced.

free parameters (7)
  • tau_s (JS divergence clean threshold) = 0.75
    Threshold in Eq. (6) for small-loss/JSD selection; set manually in Section 4.1 without sensitivity analysis.
  • tau_h (high-confidence clean threshold) = 0.9
    Threshold in Eq. (8) for including samples whose predicted probability of the annotated label exceeds 0.9; manually chosen in Section 4.1.
  • tau_p (margin threshold) = 0.9
    Threshold in Eq. (18) for selecting high-confidence ID samples in the margin-guided module; manually chosen in Section 4.1.
  • lambda_1 (noisy ID loss weight) = 0.05
    Weight on the re-labeling loss L_n in Eq. (26); set in Section 4.1.
  • lambda_2 (consistency loss weight) = 0.05
    Weight on consistency regularization L_cons in Eq. (26); set in Section 4.1.
  • epsilon (label smoothing) = not specified
    Label-smoothing parameter in Eq. (20); the paper does not report its value.
  • tau (sharpening temperature) = not specified
    Temperature in Eq. (21) for pseudo-label sharpening; value not reported.
assumptions (5)
  • domain assumption Deep networks fit clean samples before noisy ones, so low JS divergence or high confidence on the annotated label indicates a clean sample.
    Used in Section 3.1 to justify the robust sample selection module (Eqs. 5-9).
  • domain assumption OOD samples produce near-uniform predictions, so argmax disagreement between weak and strong augmented views identifies OOD samples.
    Used in Section 3.2, Eqs. (11)-(12), to partition the noisy set; not validated in the paper.
  • domain assumption ID noisy samples with margin larger than tau_p have trustworthy pseudo-labels.
    Used in Section 3.2, Eq. (18), and Section 3.3 to re-label high-confidence ID samples.
  • domain assumption The annotated labels of samples in the selected clean set D_c are correct.
    Section 3.3 treats D_c with cross-entropy on original labels; contamination would degrade training.
  • domain assumption Weak and strong augmentations of the same image share the same semantic label.
    Used in Section 3.3 for pseudo-label averaging and the consistency loss in Eq. (25).

how reviews work

0 comments
Cite this review

Pith. "Pith review of Open set label noise learning with robust sample selection and margin-guided module." pith.science (2026). https://pith.science/paper/2AUFXKIY

@misc{pith2026250104269,
  author       = {Pith},
  title        = {Pith review of: Open set label noise learning with robust sample selection and margin-guided module},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/2AUFXKIY}},
  note         = {Machine review of arXiv:2501.04269}
}
read the original abstract

In recent years, the remarkable success of deep neural networks (DNNs) in computer vision is largely due to large-scale, high-quality labeled datasets. Training directly on real-world datasets with label noise may result in overfitting. The traditional method is limited to deal with closed set label noise, where noisy training data has true class labels within the known label space. However, there are some real-world datasets containing open set label noise, which means that some samples belong to an unknown class outside the known label space. To address the open set label noise problem, we introduce a method based on Robust Sample Selection and Margin-Guided Module (RSS-MGM). Firstly, unlike the prior clean sample selection approach, which only select a limited number of clean samples, a robust sample selection module combines small loss selection or high-confidence sample selection to obtain more clean samples. Secondly, to efficiently distinguish open set label noise and closed set ones, margin functions are designed to filter open-set data and closed set data. Thirdly, different processing methods are selected for different types of samples in order to fully utilize the data's prior information and optimize the whole model. Furthermore, extensive experimental results with noisy labeled data from benchmark datasets and real-world datasets, such as CIFAR-100N-C, CIFAR80N-O, WebFG-469, and Food101N, indicate that our approach outperforms many state-of-the-art label noise learning methods. Especially, it can more accurately divide open set label noise samples and closed set ones.

Figures

Figures reproduced from arXiv: 2501.04269 by the authors.

Figure 1
Figure 1. The example of open set label noise problem. Collecting a network dataset with labels from the CIFAR10 dataset but different images. Each image in the dataset is divided into three different groups: Clean Set, Closed Set, and Open Set. Clean Set refers to images with correct labels. Closed Set signifies images that are labeled incorrectly, yet their correct labels still exist within the known label space. Open Set i… view at source ↗
Figure 2
Figure 2. The overall framework of RSS-MGM. Each input image 𝑥𝑖 is augmented into one weakly view and one strongly augmented view before being fed into the label predictor network, leading to two label predictions: 𝑝 𝑤 for the weakly augmented view and 𝑝 𝑠 for the strongly augmented view. Afterward, based on the Robust Sample Selection Module, samples are classified as Clean Set or Noisy Set. If the sample is clean, it will b… view at source ↗
Figure 3
Figure 3. An example of the small loss selection method for dividing the training dataset. Here, Clean Set represents the set of samples that are identified as clean samples, which are considered to be labeled accurately. The Noise Set, on the other hand, represents the set of samples with noisy labels. However, the samples bordered in red in the noise set are actually clean, but are incorrectly classified in the noise set. T… view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

78 extracted references · 64 canonical work pages

  1. [1]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in ":" * " " * FUNCTION f...

  2. [2]

    Krizhevsky , author I

    author A. Krizhevsky , author I. Sutskever , author G. Hinton , title Imagenet classification with deep convolutional neural networks , journal Advances in neural information processing systems volume 25 ( year 2012 )

    work page 2012

  3. [3]

    Ren , author K

    author S. Ren , author K. He , author R. Girshick , author J. Sun , title Faster r-cnn: Towards real-time object detection with region proposal networks , journal Advances in neural information processing systems volume 28 ( year 2015 )

    work page 2015

  4. [4]

    Redmon , author A

    author J. Redmon , author A. Farhadi , title Yolo9000: better, faster, stronger , in: booktitle Proceedings of the IEEE conference on computer vision and pattern recognition , year 2017 , pp. pages 7263--7271

    work page 2017

  5. [5]

    Xie , author L

    author G.-S. Xie , author L. Liu , author F. Zhu , author F. Zhao , author Z. Zhang , author Y. Yao , author J. Qin , author L. Shao , title Region graph embedding network for zero-shot learning , in: booktitle Computer Vision--ECCV 2020: 16th European Conference, Glasgow, UK, August 23--28, 2020, Proceedings, Part IV 16 , organization Springer , year 202...

    work page 2020

  6. [6]

    Luo , author G

    author H. Luo , author G. Lin , author Z. Liu , author F. Liu , author Z. Tang , author Y. Yao , title Segeqa: Video segmentation based visual attention for embodied question answering , in: booktitle 2019 IEEE/CVF International Conference on Computer Vision (ICCV) , organization IEEE , year 2019 , pp. pages 9666--9675

    work page 2019

  7. [7]

    Deng , author W

    author J. Deng , author W. Dong , author R. Socher , author L.-J. Li , author K. Li , author L. Fei-Fei , title Imagenet: A large-scale hierarchical image database , in: booktitle 2009 IEEE conference on computer vision and pattern recognition , organization Ieee , year 2009 , pp. pages 248--255

    2009

  8. [8]

    Lin , author M

    author T.-Y. Lin , author M. Maire , author S. Belongie , author J. Hays , author P. Perona , author D. Ramanan , author P. Doll \'a r , author C. L. Zitnick , title Microsoft coco: Common objects in context , in: booktitle Computer Vision--ECCV 2014: 13th European Conference, Zurich, Switzerland, September 6-12, 2014, Proceedings, Part V 13 , organizatio...

    2014

Show all 78 references
  1. [9]

    Li , author R

    author J. Li , author R. Socher , author S. C. Hoi , title Dividemix: Learning with noisy labels as semi-supervised learning , journal arXiv preprint arXiv:2002.07394 ( year 2020 )

    2002 arXiv

  2. [10]

    Wang , author R

    author H. Wang , author R. Xiao , author Y. Dong , author L. Feng , author J. Zhao , title Promix: combating label noise via maximizing clean sample utility , journal arXiv preprint arXiv:2207.10276 ( year 2022 )

    2022 arXiv

  3. [11]

    Wu , author X

    author S. Wu , author X. Xia , author T. Liu , author B. Han , author M. Gong , author N. Wang , author H. Liu , author G. Niu , title Class2simi: A new perspective on learning with label noise ( year 2020 )

    2020

  4. [12]

    Sehwag , author A

    author V. Sehwag , author A. N. Bhagoji , author L. Song , author C. Sitawarin , author D. Cullina , author M. Chiang , author P. Mittal , title Analyzing the robustness of open-world machine learning , in: booktitle Proceedings of the 12th ACM Workshop on Artificial Intellige...

    2019

  5. [13]

    Sun , author Y

    author Z. Sun , author Y. Yao , author X.-S. Wei , author Y. Zhang , author F. Shen , author J. Wu , author J. Zhang , author H. T. Shen , title Webly supervised fine-grained recognition: Benchmark datasets and an approach , in: booktitle Proceedings of the IEEE/CVF internatio...

    2021

  6. [14]

    Yang , author X

    author J. Yang , author X. Sun , author Y.-K. Lai , author L. Zheng , author M.-M. Cheng , title Recognition from web data: A progressive filtering approach , journal IEEE Transactions on Image Processing volume 27 ( year 2018 ) pages 5303--5315

    2018

  7. [15]

    Yao , author X

    author Y. Yao , author X. Hua , author G. Gao , author Z. Sun , author Z. Li , author J. Zhang , title Bridging the web data and fine-grained visual recognition via alleviating label noise and domain mismatch , in: booktitle Proceedings of the 28th ACM International Conference...

    2020

  8. [16]

    Tanaka , author D

    author D. Tanaka , author D. Ikami , author T. Yamasaki , author K. Aizawa , title Joint optimization framework for learning with noisy labels , in: booktitle Proceedings of the IEEE conference on computer vision and pattern recognition , year 2018 , pp. pages 5552--5560

    2018

  9. [17]

    Yao , author F

    author Y. Yao , author F. Shen , author J. Zhang , author L. Liu , author Z. Tang , author L. Shao , title Extracting multiple visual senses for web learning , journal IEEE Transactions on Multimedia volume 21 ( year 2018 ) pages 184--196

    2018

  10. [18]

    Zhang , author Y

    author W. Zhang , author Y. Wang , author Y. Qiao , title Metacleaner: Learning to hallucinate clean representations for noisy-labeled visual recognition , in: booktitle Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , year 2019 , pp. pages 7...

    2019

  11. [19]

    Liu , author C

    author H. Liu , author C. Zhang , author Y. Yao , author X.-S. Wei , author F. Shen , author Z. Tang , author J. Zhang , title Exploiting web images for fine-grained visual recognition by eliminating open-set noise and utilizing hard examples , journal IEEE Transactions on Mul...

    2021

  12. [20]

    Sun , author H

    author Z. Sun , author H. Liu , author Q. Wang , author T. Zhou , author Q. Wu , author Z. Tang , title Co-ldl: A co-training-based label distribution learning method for tackling label noise , journal IEEE Transactions on Multimedia volume 24 ( year 2021 ) pages 1093--1104

    2021

  13. [21]

    Li , author L

    author W. Li , author L. Wang , author W. Li , author E. Agustsson , author L. Van Gool , title Webvision database: Visual learning and understanding from web data , journal arXiv preprint arXiv:1708.02862 ( year 2017 )

    2017 arXiv

  14. [22]

    Xiao , author T

    author T. Xiao , author T. Xia , author Y. Yang , author C. Huang , author X. Wang , title Learning from massive noisy labeled data for image classification , in: booktitle Proceedings of the IEEE conference on computer vision and pattern recognition , year 2015 , pp. pages 2691--2699

    2015

  15. [23]

    Sun , author X.-S

    author Z. Sun , author X.-S. Hua , author Y. Yao , author X.-S. Wei , author G. Hu , author J. Zhang , title Crssc: salvage reusable samples from noisy data for robust learning , in: booktitle Proceedings of the 28th ACM International Conference on Multimedia , year 2020 , pp....

    2020

  16. [24]

    Yao , author J

    author Y. Yao , author J. Zhang , author F. Shen , author L. Liu , author F. Zhu , author D. Zhang , author H. T. Shen , title Towards automatic construction of diverse, high-quality image datasets , journal IEEE Transactions on Knowledge and Data Engineering volume 32 ( year ...

    2019

  17. [25]

    author D. F. Nettleton , author A. Orriols-Puig , author A. Fornells , title A study of the effect of different types of noise on the precision of supervised learning techniques , journal Artificial intelligence review volume 33 ( year 2010 ) pages 275--306

    2010

  18. [26]

    Zhang , author S

    author C. Zhang , author S. Bengio , author M. Hardt , author B. Recht , author O. Vinyals , title Understanding deep learning (still) requires rethinking generalization , journal Communications of the ACM volume 64 ( year 2021 ) pages 107--115

    2021

  19. [27]

    Arpit , author S

    author D. Arpit , author S. Jastrz e bski , author N. Ballas , author D. Krueger , author E. Bengio , author M. S. Kanwal , author T. Maharaj , author A. Fischer , author A. Courville , author Y. Bengio , et al., title A closer look at memorization in deep networks , in: bookt...

    2017

  20. [28]

    Wang , author W

    author Y. Wang , author W. Liu , author X. Ma , author J. Bailey , author H. Zha , author L. Song , author S.-T. Xia , title Iterative learning with open-set noisy labels , in: booktitle Proceedings of the IEEE conference on computer vision and pattern recognition , year 2018 ...

    2018

  21. [29]

    Lin , author P

    author T.-Y. Lin , author P. Goyal , author R. Girshick , author K. He , author P. Doll \'a r , title Focal loss for dense object detection , in: booktitle Proceedings of the IEEE international conference on computer vision , year 2017 , pp. pages 2980--2988

    2017

  22. [30]

    Ghosh , author H

    author A. Ghosh , author H. Kumar , author P. S. Sastry , title Robust loss functions under label noise for deep neural networks , in: booktitle Proceedings of the AAAI conference on artificial intelligence , volume volume 31 , year 2017

    2017

  23. [31]

    Zhang , author M

    author Z. Zhang , author M. Sabuncu , title Generalized cross entropy loss for training deep neural networks with noisy labels , journal Advances in neural information processing systems volume 31 ( year 2018 )

    2018

  24. [32]

    Wu , author R

    author X. Wu , author R. He , author Z. Sun , author T. Tan , title A light cnn for deep face representation with noisy labels , journal IEEE Transactions on Information Forensics and Security volume 13 ( year 2018 ) pages 2884--2896

    2018

  25. [33]

    Reed , author H

    author S. Reed , author H. Lee , author D. Anguelov , author C. Szegedy , author D. Erhan , author A. Rabinovich , title Training deep neural networks on noisy labels with bootstrapping , journal arXiv preprint arXiv:1412.6596 ( year 2014 )

    2014 arXiv

  26. [34]

    Arazo , author D

    author E. Arazo , author D. Ortego , author P. Albert , author N. O’Connor , author K. McGuinness , title Unsupervised label noise modeling and loss correction , in: booktitle International conference on machine learning , organization PMLR , year 2019 , pp. pages 312--321

    2019

  27. [35]

    Zheng , author P

    author S. Zheng , author P. Wu , author A. Goswami , author M. Goswami , author D. Metaxas , author C. Chen , title Error-bounded correction of noisy labels , in: booktitle International Conference on Machine Learning , organization PMLR , year 2020 , pp. pages 11447--11457

    2020

  28. [36]

    Zhang , author M

    author H. Zhang , author M. Cisse , author Y. N. Dauphin , author D. Lopez-Paz , title mixup: Beyond empirical risk minimization , journal arXiv preprint arXiv:1710.09412 ( year 2017 )

    2017 arXiv

  29. [37]

    Han , author Q

    author B. Han , author Q. Yao , author X. Yu , author G. Niu , author M. Xu , author W. Hu , author I. Tsang , author M. Sugiyama , title Co-teaching: Robust training of deep neural networks with extremely noisy labels , journal Advances in neural information processing system...

    2018

  30. [38]

    Lee , author X

    author K.-H. Lee , author X. He , author L. Zhang , author L. Yang , title Cleannet: Transfer learning for scalable image classifier training with label noise , in: booktitle Proceedings of the IEEE conference on computer vision and pattern recognition , year 2018 , pp. pages ...

    2018

  31. [39]

    Xia , author T

    author X. Xia , author T. Liu , author B. Han , author M. Gong , author J. Yu , author G. Niu , author M. Sugiyama , title Instance correction for learning with open-set noisy labels , journal arXiv preprint arXiv:2106.00455 ( year 2021 )

    2021 arXiv

  32. [40]

    Sachdeva , author F

    author R. Sachdeva , author F. R. Cordeiro , author V. Belagiannis , author I. Reid , author G. Carneiro , title Evidentialmix: Learning with combined open-set and closed-set noisy labels , in: booktitle Proceedings of the IEEE/CVF Winter Conference on Applications of Computer...

    2021

  33. [41]

    Li , author C

    author J. Li , author C. Xiong , author S. C. Hoi , title Mopro: Webly supervised learning with momentum prototypes , journal arXiv preprint arXiv:2009.07995 ( year 2020 )

    2009 arXiv

  34. [42]

    Yao , author Z

    author Y. Yao , author Z. Sun , author C. Zhang , author F. Shen , author Q. Wu , author J. Zhang , author Z. Tang , title Jo-src: A contrastive approach for combating noisy labels , in: booktitle Proceedings of the IEEE/CVF conference on computer vision and pattern recognitio...

    2021

  35. [43]

    Sun , author F

    author Z. Sun , author F. Shen , author D. Huang , author Q. Wang , author X. Shu , author Y. Yao , author J. Tang , title Pnp: Robust learning from noisy labels by probabilistic noise prediction , in: booktitle Proceedings of the IEEE/CVF Conference on Computer Vision and Pat...

    2022

  36. [44]

    Cai , author H

    author Z. Cai , author H. Liu , author D. Huang , author Y. Yao , author Z. Tang , title Co-mining: Mining informative samples with noisy labels , journal Signal Processing volume 209 ( year 2023 ) pages 109003

    2023

  37. [45]

    Albert , author E

    author P. Albert , author E. Arazo , author T. Krishna , author N. E. O’Connor , author K. McGuinness , title Is your noise correction noisy? pls: Robustness to label noise with two stage detection , in: booktitle Proceedings of the IEEE/CVF Winter Conference on Applications o...

    2023

  38. [46]

    Xu , author R

    author C. Xu , author R. Lin , author J. Cai , author S. Wang , title Label correction using contrastive prototypical classifier for noisy label learning , journal Information Sciences volume 649 ( year 2023 ) pages 119647

    2023

  39. [47]

    Yu , author B

    author X. Yu , author B. Han , author J. Yao , author G. Niu , author I. Tsang , author M. Sugiyama , title How does disagreement help generalization against label corruption? , in: booktitle International Conference on Machine Learning , organization PMLR , year 2019 , pp. pa...

    2019

  40. [48]

    Pleiss , author T

    author G. Pleiss , author T. Zhang , author E. Elenberg , author K. Q. Weinberger , title Identifying mislabeled data using the area under the margin ranking , journal Advances in Neural Information Processing Systems volume 33 ( year 2020 ) pages 17044--17056

    2020

  41. [49]

    Cheng , author Z

    author H. Cheng , author Z. Zhu , author X. Li , author Y. Gong , author X. Sun , author Y. Liu , title Learning with instance-dependent label noise: A sample sieve approach , journal arXiv preprint arXiv:2010.02347 ( year 2020 )

    2010 arXiv

  42. [50]

    Zhu , author W

    author C. Zhu , author W. Chen , author T. Peng , author Y. Wang , author M. Jin , title Hard sample aware noise robust learning for histopathology image classification , journal IEEE transactions on medical imaging volume 41 ( year 2021 ) pages 881--894

    2021

  43. [51]

    Zhang , author W

    author Y. Zhang , author W. Deng , author X. Cui , author Y. Yin , author H. Shi , author D. Wen , title Model and data agreement for learning with noisy labels , journal arXiv preprint arXiv:2212.01054 ( year 2022 )

    2022 arXiv

  44. [52]

    author F. R. Cordeiro , author R. Sachdeva , author V. Belagiannis , author I. Reid , author G. Carneiro , title Longremix: Robust learning with high confidence samples in a noisy label environment , journal Pattern Recognition volume 133 ( year 2023 ) pages 109013

    2023

  45. [53]

    Wen , author H

    author Z. Wen , author H. Xu , author S. Ying , title Jsmix: a holistic algorithm for learning with label noise , journal Neural Computing and Applications volume 35 ( year 2023 ) pages 1519--1533

    2023

  46. [54]

    Li , author Z

    author Y. Li , author Z. Guo , author L. Wang , author L. Xu , title Tbc-mi : Suppressing noise labels by maximizing cleaning samples for robust image classification , journal Information Processing & Management volume 61 ( year 2024 ) pages 103801 . https://www.sciencedirect....

    2024

  47. [55]

    Wan , author X

    author W. Wan , author X. Wang , author M.-K. Xie , author S.-Y. Li , author S.-J. Huang , author S. Chen , title Unlocking the power of open set: A new perspective for open-set noisy label learning , in: booktitle Proceedings of the AAAI conference on artificial intelligence ...

    2024

  48. [56]

    Chen , author S

    author T. Chen , author S. Kornblith , author M. Norouzi , author G. Hinton , title A simple framework for contrastive learning of visual representations , in: booktitle International conference on machine learning , organization PMLR , year 2020 a , pp. pages 1597--1607

    2020

  49. [57]

    Chen , author H

    author X. Chen , author H. Fan , author R. Girshick , author K. He , title Improved baselines with momentum contrastive learning , journal arXiv preprint arXiv:2003.04297 ( year 2020 b )

    2003 arXiv

  50. [58]

    Ortego , author E

    author D. Ortego , author E. Arazo , author P. Albert , author N. E. O'Connor , author K. McGuinness , title Multi-objective interpolation training for robustness to label noise , in: booktitle Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition ,...

    2021

  51. [59]

    Yang , author Y

    author J. Yang , author Y. Bisk , author J. Gao , title Taco: Token-aware cascade contrastive learning for video-text alignment , in: booktitle Proceedings of the IEEE/CVF International Conference on Computer Vision , year 2021 , pp. pages 11562--11572

    2021

  52. [60]

    Li , author C

    author J. Li , author C. Xiong , author S. C. Hoi , title Learning from noisy data with robust representation learning , in: booktitle Proceedings of the IEEE/CVF International Conference on Computer Vision , year 2021 , pp. pages 9485--9494

    2021

  53. [61]

    Jiang , author D

    author L. Jiang , author D. Huang , author M. Liu , author W. Yang , title Beyond synthetic noise: Deep learning on controlled noisy labels , in: booktitle International conference on machine learning , organization PMLR , year 2020 , pp. pages 4804--4815

    2020

  54. [62]

    Yan , author L

    author J. Yan , author L. Luo , author C. Xu , author C. Deng , author H. Huang , title Noise is also useful: Negative correlation-steered latent contrastive learning , in: booktitle Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , year 2022 ...

    2022

  55. [63]

    Zhang , author S

    author C. Zhang , author S. Bengio , author M. Hardt , author B. Recht , author O. Vinyals , title Understanding deep learning requires rethinking generalization , in: booktitle 5th International Conference on Learning Representations, ICLR 2017, Toulon, France, April 24-26, 2...

    2017

  56. [64]

    Sohn , author D

    author K. Sohn , author D. Berthelot , author N. Carlini , author Z. Zhang , author H. Zhang , author C. A. Raffel , author E. D. Cubuk , author A. Kurakin , author C.-L. Li , title Fixmatch: Simplifying semi-supervised learning with consistency and confidence , journal Advanc...

    2020

  57. [65]

    Zhao , author L

    author Z. Zhao , author L. Zhou , author L. Wang , author Y. Shi , author Y. Gao , title Lassl: Label-guided self-training for semi-supervised learning , in: booktitle Proceedings of the AAAI conference on artificial intelligence , volume volume 36 , year 2022 , pp. pages 9208--9216

    2022

  58. [66]

    Wang , author R

    author L. Wang , author R. Chan , author T. Zeng , title Probabilistic semi-supervised learning via sparse graph structure learning , journal IEEE transactions on neural networks and learning systems volume 32 ( year 2020 ) pages 853--867

    2020

  59. [67]

    Gu , author M

    author N. Gu , author M. Fan , author D. Meng , title Robust semi-supervised classification for noisy labels based on self-paced learning , journal IEEE Signal Processing Letters volume 23 ( year 2016 ) pages 1806--1810

    2016

  60. [68]

    Zhang , author D

    author L. Zhang , author D. Xie , author Y. Li , author A. Camargo , author D. Song , author T. Lu , author J. Jeudy , author D. Dreizin , author E. R. Melhem , author Z. Wang , et al., title Improving sensitivity of arterial spin labeling perfusion mri in alzheimer's disease ...

    2022

  61. [69]

    Krizhevsky , author G

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

    2009

  62. [70]

    Bossard , author M

    author L. Bossard , author M. Guillaumin , author L. Van Gool , title Food-101--mining discriminative components with random forests , in: booktitle Computer Vision--ECCV 2014: 13th European Conference, Zurich, Switzerland, September 6-12, 2014, Proceedings, Part VI 13 , organ...

    2014

  63. [71]

    when to update

    author E. Malach , author S. Shalev-Shwartz , title Decoupling" when to update" from" how to update" , journal Advances in neural information processing systems volume 30 ( year 2017 )

    2017

  64. [72]

    Wei , author L

    author H. Wei , author L. Feng , author X. Chen , author B. An , title Combating noisy labels by agreement: A joint training method with co-regularization , in: booktitle Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , year 2020 , pp. pages ...

    2020

  65. [73]

    Song , author M

    author H. Song , author M. Kim , author J.-G. Lee , title Selfie: Refurbishing unclean samples for robust deep learning , in: booktitle International Conference on Machine Learning , organization PMLR , year 2019 , pp. pages 5907--5915

    2019

  66. [74]

    Yi , author J

    author K. Yi , author J. Wu , title Probabilistic end-to-end noise correction for learning with noisy labels , in: booktitle Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , year 2019 , pp. pages 7017--7025

    2019

  67. [75]

    Peng , author K

    author X. Peng , author K. Wang , author Z. Zeng , author Q. Li , author J. Yang , author Y. Qiao , title Suppressing mislabeled data via grouping and self-attention , in: booktitle Computer Vision--ECCV 2020: 16th European Conference, Glasgow, UK, August 23--28, 2020, Proceed...

    2020

  68. [76]

    Huang , author C

    author L. Huang , author C. Zhang , author H. Zhang , title Self-adaptive training: beyond empirical risk minimization , journal Advances in neural information processing systems volume 33 ( year 2020 ) pages 19365--19376

    2020

  69. [77]

    Zhang , author S

    author Y. Zhang , author S. Zheng , author P. Wu , author M. Goswami , author C. Chen , title Learning with feature-dependent label noise: A progressive approach , journal arXiv preprint arXiv:2103.07756 ( year 2021 )

    2021 arXiv

  70. [78]

    Han , author P

    author J. Han , author P. Luo , author X. Wang , title Deep self-learning from noisy labels , in: booktitle Proceedings of the IEEE/CVF international conference on computer vision , year 2019 , pp. pages 5138--5147

    2019

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.