Pith. sign in

REVIEW 4 major objections 7 minor 29 references

LiNC: Lightweight Noise Correction via Per-Sample Trust and Gaussian Mixture Modeling

T0 review · 4 major / 7 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read A per-sample trust parameter learned during ordinary training separates clean from noisy labels, and a Gaussian mixture model then corrects only the low-trust samples.

desk verdict Useful, lightweight label-noise method with a clean trust-gradient derivation, but the alpha weight-decay term and missing error bars undermine the central separability claim; worth refining and re-reviewing. read the letter →

arxiv 2608.04147 v1 pith:GVDXOO3Q submitted 2026-08-04 cs.LG cs.AIcs.CV

Abhishek Moturu , Babak Taati , Anna Goldenberg This is my paper · ORCID
classification cs.LGcs.AIcs.CV
keywords labelnoisenoisylabelsper-sampletrustGaussianmixturemodelcorrectionmedicalimagingMedMNISTearlylearning
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

LiNC claims that a single trainable trust parameter per training sample can separate correct from corrupted labels during the first epochs of ordinary training, without an extra model, a clean validation set, or a manually chosen threshold. The paper derives the gradient of the training objective with respect to that trust parameter and shows it pushes trust upward when the model agrees with the observed label and downward when it disagrees. A three-component Gaussian mixture model then groups trust values into noisy, ambiguous, and clean cases, and only the noisy group is relabeled, first softly and then hard. Across ten 2D medical imaging datasets with symmetric label noise up to 50%, the method raises mean last-epoch accuracy by 21.41 percentage points at the highest noise rate and provides a per-sample trust score that can also be used to audit datasets.

What carries the argument

The load-bearing object is the per-sample trust parameter $\alpha_i\in(0,1)$, which interpolates between the observed one-hot label and the model's own prediction through the soft target $q_i(\alpha_i)=(1-\alpha_i)\,\mathrm{stopgrad}(p_i)+\alpha_i e_{\tilde y_i}$. The argument turns on the trust-gradient identity $\partial L_i/\partial\alpha_i = -\log p_{i,\tilde y_i}-H(p_i)$: in early training the model is expected to fit correct patterns first, so the sign of this gradient separates clean from noisy samples. The second mechanism is a fixed three-component Gaussian mixture model fitted to the trust values, which provides a threshold-free partition into noisy, ambiguous, and clean groups; only the noisy component is later corrected, first with soft targets and then with hard labels.

What would settle it

Train LiNC on a dataset where one class has all its labels systematically flipped to a single incorrect class; if those samples receive high trust because the model confidently agrees with the wrong label, the class-conditional correction rate and accuracy will reveal the failure. A second test is to run LiNC on a fully clean dataset and check whether the fixed three-component GMM still sends a nonzero fraction of true labels to hard correction.

Watch

Extended reading notes

Core claim

The paper's central claim is that the sign of the trust gradient is determined by a comparison between the negative log-likelihood of the observed label, $-\log p_{i,\tilde y_i}$, and the prediction entropy $H(p_i)$: during early training, $-\log p_{i,\tilde y_i}<H(p_i)$ for clean samples, so $\alpha_i$ rises, while $-\log p_{i,\tilde y_i}>H(p_i)$ for mislabeled samples, so $\alpha_i$ falls. This produces a separable distribution of trust values without access to ground-truth labels. Fitting a three-component Gaussian mixture model to the trust values, LiNC assigns the lowest-mean component to noisy samples, trains those samples first with the model's soft predictions and then with the hard $\arg\max$ label, and leaves the other samples untouched. On ten MedMNISTv2 datasets under symmetric noise, the method reports mean last-epoch accuracy gains of 2.19, 7.86, 12.31, 16.94, and 21.41 percentage points at noise rates of 10% through 50%, and improves mislabel-detection AUC to 0.9837 on OrganSMNIST at 20% noise.

Load-bearing premise

The method rests on the empirical early-learning premise that, in the first epochs, the model assigns higher probability to correct labels than to incorrect ones; if a wrong label is systematically repeated or a correct label is hard, the trust gradient can point the wrong way and the correction will miss or damage those samples.

Editorial extensions

If this is right

  • At 50% symmetric label noise, mean last-epoch accuracy across the ten datasets rises from 0.5882 to 0.8023, a gain of 21.41 percentage points, and LiNC beats standard training in 49 of 50 dataset-noise combinations on last-epoch accuracy.
  • The gap between best-epoch and last-epoch accuracy shrinks from 13.14 to 2.31 percentage points on average, which means the method suppresses late-training memorization of corrupted labels.
  • At 20% noise on OrganSMNIST, the trust values detect mislabeled samples with AUC 0.9837, better than the strongest listed baseline at 0.9126, so the same learned scores can be used for dataset auditing.
  • The method adds $O(N)$ memory for trust parameters and keeps the asymptotic training-time complexity at $O(EN)$, so it can be added to a standard fine-tuning loop.
  • The staged correction schedule, soft warmup, soft correction, then hard correction, limits the risk of reinforcing wrong model predictions when the model itself is still unreliable.

Reading between the lines

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

  • A testable extension of this mechanism is that any loss whose gradient separates agreement from disagreement could generate similar trust scores, so the approach may transfer beyond cross-entropy to other training objectives.
  • Because the separation depends on the early-learning assumption, systematically corrupted subgroups, such as a rare class whose instances all receive the same wrong label, could keep high trust and evade correction; the paper itself flags this risk.
  • The fixed three-component GMM will always find three groups even on clean data, and the paper's PathMNIST 10% result (last-epoch accuracy 0.9154 with LiNC versus 0.9536 without) is consistent with over-correction, suggesting a data-driven component count or an abstention rule would help.
  • The trust scores could be reused after training as a ranking for expert re-review, subgroup fairness analysis, or sample reweighting, even in settings where label correction itself is not desired.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 7 minor

Summary. The paper proposes LiNC, a method that augments standard supervised training with a per-sample trainable trust parameter α_i. The training target q_i is a convex combination of the observed one-hot label and the model's own predictive distribution, q_i = (1−α_i) stopgrad(p_i) + α_i e_{y_i}. The authors derive the gradient of the loss with respect to α_i (Eq. 5) and argue from the early-learning phenomenon that this gradient pushes α_i upward for clean samples (where the negative log-likelihood of the observed label is below the prediction entropy) and downward for mislabeled samples (where it is above). After a five-epoch warmup, a 3-component Gaussian Mixture Model is fit to the trust values, and samples in the lowest-mean component are corrected, first via soft targets from the model and later via hard argmax labels. Experiments on ten MedMNISTv2 datasets with symmetric label noise up to 50% report mean last-epoch accuracy improvements of 21.41 percentage points at 50% noise relative to standard training, AUC 0.9837 for mislabel detection on OrganSMNIST, and O(N) additional memory with negligible asymptotic runtime overhead.

Significance. If the empirical claims hold, LiNC is attractive because it is conceptually simple, requires no second network, no clean validation set, and no manually tuned threshold, and it produces a per-sample trust score that could be used for dataset auditing. The gradient identity in Theorem 3.1 is correctly derived, and Figure 2 gives visual evidence that the trust values separate under symmetric noise on OrganSMNIST. The method's asymptotic overhead is genuinely small, and the authors are explicit about several limitations, including the rare-class risk and the fixed K=3 GMM. However, the strength of the conclusions is limited by the experimental protocol (single runs, no error bars), the single-dataset detection comparison, and a theoretical analysis that omits the effect of the trust weight decay. These issues are load-bearing for the headline claims, so the paper currently falls short of the standard for acceptance, but the core idea is promising and worth revising.

major comments (4)
  1. [§3.3, Eq. (5); Algorithm 1 line 19] The sign analysis of the trust gradient considers only ∂L/∂α_i = NLL − H, but the actual trust update in Algorithm 1 line 19 includes weight decay: α_i ← α_i − η_α(NLL − H + λ_α α_i). With η_α=1 and λ_α=0.1, the fixed point is NLL − H = −0.1 α_i, not NLL − H = 0. Consequently, a clean sample with NLL − H slightly negative (but larger than −0.1 α_i) will have its trust decay at every optimizer step, even though the paper's clean-label condition NLL < H is satisfied. Over the five-epoch warmup, this can push hard-but-clean samples into the low-trust GMM component and cause them to be hard-corrected to the model's current argmax, which may be wrong. This provides a plausible mechanism for the PathMNIST 10% noise result in Table 3, where LiNC's last-epoch accuracy (0.9154) is 3.8 points below standard training (0.9536). The manuscript should analyze the equilibrium including λ_α, report the sensitivity of the results to λ_α, or modify the trust update so that the clean/noise boundary is not shifted.
  2. [§4.2, Table 3] All numerical results are reported as single runs with no repeated seeds, standard deviations, or confidence intervals. The headline improvement of 21.41 percentage points in mean last-epoch accuracy at 50% noise is a point estimate from one training run per configuration; on the smaller MedMNISTv2 datasets, such as BreastMNIST and PneumoniaMNIST, single-seed accuracy differences can easily exceed several points. The conclusions 'LiNC achieves higher last-epoch accuracy in 49 of the 50 dataset-noise combinations' and 'consistent gains' are therefore not supported by the evidence as reported. Please provide means and standard deviations over at least three seeds, or justify that the experimental setting is deterministic and that the reported numbers are stable.
  3. [§4.3, Table 1] The mislabel detection comparison in Table 1 is restricted to a single dataset (OrganSMNIST at 20% noise) and provides no experimental details for the seven baselines. The paper does not state how AUM, DataMaps, EL2N, Forgetting, CNLCU-S, or V oG were implemented, which underlying scores they used, how hyperparameters were selected, or whether the reported AUC values are averaged over runs. Without this information, the claimed improvement (LiNC 0.9837 vs V oG 0.9126) is not reproducible. Please add implementation details for all baselines, report results on more than one dataset, and include error bars or at least multiple runs to substantiate the superiority claim.
  4. [§3.5, Algorithm 1 lines 10–14; §5] The hard-correction stage replaces the label of every sample in the lowest-trust GMM component with the model's current argmax, regardless of the model's confidence on that sample. If a non-negligible fraction of that component is actually clean—as the weight-decay equilibrium discussed in the first major comment suggests—then the method is actively replacing correct labels with incorrect ones. The paper should quantify this risk by reporting the precision and recall of the GMM 'noisy' component relative to the true injected noise, as well as the accuracy of the new hard labels compared to the original labels. This analysis is necessary to demonstrate that the correction step is net beneficial, and it would also clarify the PathMNIST 10% failure case that the authors themselves flag in Section 5.
minor comments (7)
  1. [§3.3, Theorem 3.1] The theorem statement says that 'if the model agrees with the observed label, then ∂L/∂α_i < 0 and if the model disagrees ... ∂L/∂α_i > 0.' This is not a mathematical consequence of the gradient formula alone; it follows only under the empirical early-training assumption that NLL < H for clean samples and NLL > H for mislabeled samples. Please rephrase the theorem to state the gradient identity and the sign condition separately, and make explicit that the clean/noisy mapping rests on the early-learning hypothesis.
  2. [§3.3, Eqs. (4) and (5)] Equations (4) and (5) are redundant derivations of the same identity; keeping only one of them would reduce clutter without changing the content.
  3. [Figure 2] In Figure 2, overlaying the GMM decision boundaries and reporting the fraction of true clean samples assigned to the noisy component would make the visual claim of separation more quantitative and easier to check.
  4. [§4.2, Table 2] Table 2 reproduces baseline numbers from reference [26] and not from experiments in this paper. Please state this explicitly in the caption and text so readers do not mistakenly attribute those results to the authors' own runs.
  5. [§4.2] The sentence 'The optimal hyperparameters without LiNC are still optimal with LiNC' is made without any ablation or sensitivity study. Either add supporting evidence or soften the claim.
  6. [Algorithm 1 line 26] After the hard-correction step, the trust parameters are frozen. The paper should briefly explain why α is no longer updated or used, since the method description earlier implies continuous trust learning.
  7. [§5] Given that the fixed K=3 GMM splits into three components even when the data are clean or lightly noisy, the paper's claim of 'consistent gains' is strained by the PathMNIST 10% result. A simple abstention rule (e.g., skip hard correction when the trust distribution is unimodal) might mitigate this, and an experiment along these lines would strengthen the discussion in Section 5.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the central derivation is a conditional calculus identity plus an external early-learning assumption, and the main empirical claims are evaluated against external benchmarks and injected noise labels.

full rationale

LiNC's derivation chain is not circular. Theorem 3.1 computes ∂L_i/∂α_i = −log p_i,tilde{y}_i − H(p_i) (Eq. 5) by direct differentiation of Eq. (2); the claimed clean/noisy separation is conditional on the early-learning behavior of deep networks, which is cited to external work (Arpit et al. [2]), not assumed through LiNC's own definitions. The GMM is fitted to the learned α values without access to the synthetic noise labels, and the reported AUC in Table 1 measures agreement with those externally injected labels, so the detection result is not a fitted parameter renamed as a prediction. Accuracy gains are measured against held-out MedMNISTv2 labels and compared with the same model without LiNC in Table 3, so the central empirical claims are externally anchored. The method is self-referential in the broad sense that model predictions become soft and hard targets (Eq. (1), Algorithm 1 lines 11 and 26), but this is an explicit design choice mitigated by stopgrad and staged correction, and the evaluation does not reduce to the model's own outputs. Section 5 candidly lists limitations, including the fixed three-component GMM and the reliance on early-learning behavior; these weaken generality but do not make the derivation circular. There are no load-bearing self-citations or imported uniqueness theorems. Concerns about the α-update's weight-decay equilibrium or the fixed GMM are correctness and robustness issues, not circularity, and the paper itself flags the GMM issue in Section 5. Therefore no circular step is identified.

Assumptions & free parameters 5 free parameters · 3 assumptions · 1 invented entities

The central claim rests on free hyperparameters for the trust update and GMM, an empirical early-learning assumption, and a fixed GMM shape. The only invented entity is the per-sample trust scalar, which is trained and validated rather than postulated without evidence.

free parameters (5)
  • Trust learning rate eta_alpha = 1
    Chosen by hand in Section 4.2; controls the size of alpha updates.
  • Trust weight decay lambda_alpha = 1e-1
    Chosen by hand in Section 4.2; regularizes the trust parameters.
  • Soft warmup epochs w = 5
    Set in Section 4.2; the authors state performance is relatively insensitive but provide no sensitivity analysis.
  • Soft correction epochs s = 5
    Set in Section 4.2; same lack of sensitivity evidence.
  • Number of GMM components K = 3
    Chosen in Section 3.4 because K=2 over-corrected borderline cases in preliminary experiments.
assumptions (3)
  • domain assumption During early training, deep networks learn clean patterns before memorizing noise, and for mislabeled samples the observed label's negative log-likelihood exceeds the model's entropy.
    Used in Theorem 3.1 to conclude trust values move in opposite directions for clean versus noisy samples; cited from Arpit et al. [2] and acknowledged as a limitation in Section 5.
  • domain assumption The distribution of trust values across samples is well approximated by a 3-component Gaussian mixture.
    Section 3.4 fixes K=3 for all datasets; the authors note in Section 5 that the GMM returns three components even when no label corruption exists.
  • standard math Standard EM on the trust scalars converges to components ordered by mean, allowing the noisy/ambiguous/clean assignment.
    Invoked in Section 3.4 for the GMM fitting; assumes the mixture components are identifiable by their means.
invented entities (1)
  • Per-sample trust parameter alpha_i independent evidence
    purpose: Controls the convex combination of the observed label and the model's own prediction in the training target (Eq. 1).
    Figure 2 shows alpha separates clean and noisy samples, and Table 1 reports AUC 0.9837 for mislabel detection; the ranking can be checked against true labels in any dataset with known corruption, so it is a falsifiable handle.

how reviews work

0 comments
Cite this review

Pith. "Pith review of LiNC: Lightweight Noise Correction via Per-Sample Trust and Gaussian Mixture Modeling." pith.science (2026). https://pith.science/paper/GVDXOO3Q

@misc{pith2026260804147,
  author       = {Pith},
  title        = {Pith review of: LiNC: Lightweight Noise Correction via Per-Sample Trust and Gaussian Mixture Modeling},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/GVDXOO3Q}},
  note         = {Machine review of arXiv:2608.04147}
}
read the original abstract

Label noise is common in medical imaging datasets due to factors such as inter-rater variability, annotation errors, and ambiguous cases. This can severely undermine the reliability and clinical effectiveness of machine learning models trained using those datasets. To address this challenge, we introduce Lightweight Noise Correction (LiNC), which adds a single trainable trust parameter per training sample and learns when to use the observed label and when to defer to the model during a standard training loop. The key idea is to train using a convex combination of the observed label and the model's own predictive distribution, controlled by a per-sample trust parameter. We show that the gradient of this objective drives trust values in opposite directions for clean versus noisy samples in the early training phase, yielding separable trust distributions. We use a 3-component Gaussian Mixture Model over the trust values to separate them into clean, ambiguous, and noisy cases and then execute a short soft-correction phase on the noisy cases and a final hard correction phase. Experiments on ten 2D datasets from MedMNISTv2 under label noise of up to 50% show consistent gains in accuracy and strong mislabel detection. LiNC adds negligible asymptotic overhead: the training-time complexity remains dominated by the base network, with additional memory growing linearly with the size of the training set.

Figures

Figures reproduced from arXiv: 2608.04147 by the authors.

Figure 1
Figure 1. Overview of LiNC. The model produces a class-probability vector pi for each training sample. During soft warm-up, LiNC jointly learns a per-sample trust parameter αi and trains using the target qi = (1−αi) stopgrad(pi) +αiey˜i , which interpolates between the model prediction and the observed label y˜i . A three-component Gaussian mixture model (3-GMM) is then fitted to the learned trust values, with components orde… view at source ↗
Figure 2
Figure 2. Empirical separability of trust values αi on OrganSMNIST under 10-50% symmetric noise. Green / pink are true clean / noisy subsets, respectively, and the dashed lines indicate GMM component means (noisy / middle / clean). 5 [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Test top-1 accuracy over 70 epochs for DermaMNIST with 50% symmetric noise. LiNC [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

29 extracted references · 16 canonical work pages

  1. [1]

    Estimating example difficulty using vari- ance of gradients, 2022

    Chirag Agarwal, Daniel D’souza, and Sara Hooker. Estimating example difficulty using vari- ance of gradients, 2022

    work page 2022

  2. [2]

    A closer look at memorization in deep networks

    Devansh Arpit, Stanisław Jastrz˛ ebski, Nicolas Ballas, David Krueger, Emmanuel Bengio, Maxinder S Kanwal, Tegan Maharaj, Asja Fischer, Aaron Courville, Yoshua Bengio, et al. A closer look at memorization in deep networks. InInternational conference on machine learning, pages 233–242. PMLR, 2017

    work page 2017

  3. [3]

    Mixmatch: A holistic approach to semi-supervised learning.Advances in neural information processing systems, 32, 2019

    David Berthelot, Nicholas Carlini, Ian Goodfellow, Nicolas Papernot, Avital Oliver, and Colin A Raffel. Mixmatch: A holistic approach to semi-supervised learning.Advances in neural information processing systems, 32, 2019

    2019

  4. [4]

    Maximum likelihood from incomplete data via the em algorithm.Journal of the royal statistical society: series B (methodological), 39(1):1–22, 1977

    Arthur P Dempster, Nan M Laird, and Donald B Rubin. Maximum likelihood from incomplete data via the em algorithm.Journal of the royal statistical society: series B (methodological), 39(1):1–22, 1977

    1977

  5. [5]

    An image is worth 16x16 words: Transformers for image recognition at scale.arXiv preprint arXiv:2010.11929, 2020

    Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, et al. An image is worth 16x16 words: Transformers for image recognition at scale.arXiv preprint arXiv:2010.11929, 2020

    arXiv 2010

  6. [6]

    Robust loss functions under label noise for deep neural networks

    Aritra Ghosh, Himanshu Kumar, and P Shanti Sastry. Robust loss functions under label noise for deep neural networks. InProceedings of the AAAI conference on artificial intelligence, volume 31, 2017

    2017

  7. [7]

    Training deep neural-networks using a noise adap- tation layer

    Jacob Goldberger and Ehud Ben-Reuven. Training deep neural-networks using a noise adap- tation layer. InInternational conference on learning representations, 2017

    work page 2017

  8. [8]

    Weinberger

    Chuan Guo, Geoff Pleiss, Yu Sun, and Kilian Q. Weinberger. On calibration of modern neural networks, 2017

    2017

Show all 29 references
  1. [9]

    Co-teaching: Robust training of deep neural networks with extremely noisy labels.Advances in neural information processing systems, 31, 2018

    Bo Han, Quanming Yao, Xingrui Yu, Gang Niu, Miao Xu, Weihua Hu, Ivor Tsang, and Masashi Sugiyama. Co-teaching: Robust training of deep neural networks with extremely noisy labels.Advances in neural information processing systems, 31, 2018

    2018

  2. [10]

    Deep learning with noisy labels: Exploring techniques and remedies in medical image analysis.Medical image analysis, 65:101759, 2020

    Davood Karimi, Haoran Dou, Simon K Warfield, and Ali Gholipour. Deep learning with noisy labels: Exploring techniques and remedies in medical image analysis.Medical image analysis, 65:101759, 2020

    2020

  3. [11]

    Adam: A method for stochastic optimization.arXiv preprint arXiv:1412.6980, 2014

    Diederik P Kingma. Adam: A method for stochastic optimization.arXiv preprint arXiv:1412.6980, 2014

    2014 arXiv

  4. [12]

    Pseudo-label: The simple and efficient semi-supervised learning method for deep neural networks, 2013

    Dong-Hyun Lee et al. Pseudo-label: The simple and efficient semi-supervised learning method for deep neural networks, 2013

    2013

  5. [13]

    Dividemix: Learning with noisy labels as semi-supervised learning.arXiv preprint arXiv:2002.07394, 2020

    Junnan Li, Richard Socher, and Steven CH Hoi. Dividemix: Learning with noisy labels as semi-supervised learning.arXiv preprint arXiv:2002.07394, 2020. 10

    2002 arXiv

  6. [14]

    Making deep neural networks robust to label noise: A loss correction approach

    Giorgio Patrini, Alessandro Rozza, Aditya Krishna Menon, Richard Nock, and Lizhen Qu. Making deep neural networks robust to label noise: A loss correction approach. InProceedings of the IEEE conference on computer vision and pattern recognition, pages 1944–1952, 2017

    1944

  7. [15]

    Deep learning on a data diet: Finding important examples early in training, 2023

    Mansheej Paul, Surya Ganguli, and Gintare Karolina Dziugaite. Deep learning on a data diet: Finding important examples early in training, 2023

    2023

  8. [16]

    Identifying mislabeled data using the area under the margin ranking

    Geoff Pleiss, Tianyi Zhang, Ethan Elenberg, and Kilian Q Weinberger. Identifying mislabeled data using the area under the margin ranking. In H. Larochelle, M. Ranzato, R. Hadsell, M.F. Balcan, and H. Lin, editors,Advances in Neural Information Processing Systems, volume 33, pa...

    2020

  9. [17]

    Training deep neural networks on noisy labels with bootstrapping.arXiv preprint arXiv:1412.6596, 2014

    Scott Reed, Honglak Lee, Dragomir Anguelov, Christian Szegedy, Dumitru Erhan, and An- drew Rabinovich. Training deep neural networks on noisy labels with bootstrapping.arXiv preprint arXiv:1412.6596, 2014

    2014 arXiv

  10. [18]

    Dissecting sample hardness: A fine-grained analysis of hardness characterization methods for data-centric AI, 2024

    Nabeel Seedat, Fergus Imrie, and Mihaela van der Schaar. Dissecting sample hardness: A fine-grained analysis of hardness characterization methods for data-centric AI, 2024

    2024

  11. [19]

    Selfie: Refurbishing unclean samples for robust deep learning

    Hwanjun Song, Minseok Kim, and Jae-Gil Lee. Selfie: Refurbishing unclean samples for robust deep learning. InInternational conference on machine learning, pages 5907–5915. PMLR, 2019

    2019

  12. [20]

    Smith, and Yejin Choi

    Swabha Swayamdipta, Roy Schwartz, Nicholas Lourie, Yizhong Wang, Hannaneh Hajishirzi, Noah A. Smith, and Yejin Choi. Dataset cartography: Mapping and diagnosing datasets with training dynamics, 2020

    2020

  13. [21]

    Mariya Toneva, Alessandro Sordoni, Remi Tachet des Combes, Adam Trischler, Yoshua Ben- gio, and Geoffrey J. Gordon. An empirical study of example forgetting during deep neural network learning, 2019

    2019

  14. [22]

    Proselflc: Progressive self label correction for training robust deep neural networks

    Xinshao Wang, Yang Hua, Elyor Kodirov, David A Clifton, and Neil M Robertson. Proselflc: Progressive self label correction for training robust deep neural networks. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 752–761, 2021

    2021

  15. [23]

    Symmetric cross entropy for robust learning with noisy labels

    Yisen Wang, Xingjun Ma, Zaiyi Chen, Yuan Luo, Jinfeng Yi, and James Bailey. Symmetric cross entropy for robust learning with noisy labels. InProceedings of the IEEE/CVF interna- tional conference on computer vision, pages 322–330, 2019

    2019

  16. [24]

    Combating noisy labels by agreement: A joint training method with co-regularization

    Hongxin Wei, Lei Feng, Xiangyu Chen, and Bo An. Combating noisy labels by agreement: A joint training method with co-regularization. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 13726–13735, 2020

    2020

  17. [25]

    Sample selection with uncertainty of losses for learning with noisy labels, 2021

    Xiaobo Xia, Tongliang Liu, Bo Han, Mingming Gong, Jun Yu, Gang Niu, and Masashi Sugiyama. Sample selection with uncertainty of losses for learning with noisy labels, 2021

    2021

  18. [26]

    Medmnist v2-a large-scale lightweight benchmark for 2d and 3d biomedical image classification.Scientific Data, 10(1):41, 2023

    Jiancheng Yang, Rui Shi, Donglai Wei, Zequan Liu, Lin Zhao, Bilian Ke, Hanspeter Pfister, and Bingbing Ni. Medmnist v2-a large-scale lightweight benchmark for 2d and 3d biomedical image classification.Scientific Data, 10(1):41, 2023

    2023

  19. [27]

    Under- standing deep learning requires rethinking generalization.arXiv preprint arXiv:1611.03530, 2016

    Chiyuan Zhang, Samy Bengio, Moritz Hardt, Benjamin Recht, and Oriol Vinyals. Under- standing deep learning requires rethinking generalization.arXiv preprint arXiv:1611.03530, 2016

    2016 arXiv

  20. [28]

    Understand- ing deep learning (still) requires rethinking generalization.Communications of the ACM, 64 (3):107–115, 2021

    Chiyuan Zhang, Samy Bengio, Moritz Hardt, Benjamin Recht, and Oriol Vinyals. Understand- ing deep learning (still) requires rethinking generalization.Communications of the ACM, 64 (3):107–115, 2021

    2021

  21. [29]

    Generalized cross entropy loss for training deep neural net- works with noisy labels.Advances in neural information processing systems, 31, 2018

    Zhilu Zhang and Mert Sabuncu. Generalized cross entropy loss for training deep neural net- works with noisy labels.Advances in neural information processing systems, 31, 2018. 11

    2018

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.