pith. sign in

arxiv: 2606.29837 · v1 · pith:5MMRIMHNnew · submitted 2026-06-29 · 💻 cs.CV

Robust Trajectory Distillation: Hybrid Reweighting Meets Teacher-Inspired Targets

Kaifeng Chen , Lechao Cheng , Jiyang Li , Shengeng Tang , Fan Zhang , Yantao Pan , Yaxiong Wang , Tuanrui Hui
show 1 more author
Zhun Zhong
This is my paper

Pith reviewed 2026-06-30 06:51 UTC · model grok-4.3

classification 💻 cs.CV
keywords dataset distillationnoisy labelstrajectory distillationreweightingteacher guidancerobust learninglabel noise
0
0 comments X

The pith

A trajectory-based distillation method reweights samples by forgetting patterns and adds teacher-derived auxiliary targets to handle noisy labels without clean data.

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

Dataset distillation creates compact subsets from large labeled collections for fast training. Noisy labels can cause these subsets to embed errors, and prior fixes often demand clean examples or joint optimization that clashes with distillation goals. The paper introduces a framework that follows the teacher model's training path to downweight likely noisy items via a mix of global forgetting signals and local consistency checks. It further supplies auxiliary targets drawn from the teacher's intermediate states to strengthen useful patterns. This produces smaller datasets that train models more reliably across different noise patterns while keeping original labels intact.

Core claim

The paper establishes that Selective Guidance Reweighting fuses second-split forgetting patterns with neighborhood consistency to progressively prioritize clean supervision along the teacher trajectory, while Teacher-Inspired Auxiliary Targets supply residual guidance from intermediate teacher dynamics; together these components yield distilled datasets whose representations remain cleaner and more informative under noisy supervision without relabeling or clean anchors.

What carries the argument

Selective Guidance Reweighting (SGR) combined with Teacher-Inspired Auxiliary Targets (TIAT) applied to the teacher trajectory.

If this is right

  • Distilled subsets preserve more transferable knowledge even when original labels contain symmetric or asymmetric noise.
  • The approach remains effective on real-world noisy collections without needing clean reference data.
  • Training costs stay low because the method adds only lightweight reweighting and auxiliary signals during distillation.
  • Original labels are kept unchanged, avoiding confirmation bias from iterative correction steps.

Where Pith is reading between the lines

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

  • The trajectory analysis could be adapted to distill from web-scale scraped data where noise is common but unknown.
  • Similar forgetting-based reweighting might improve robustness in continual learning or federated settings with label noise.
  • Testing the method on non-image modalities would show whether trajectory reweighting generalizes beyond vision tasks.

Load-bearing premise

Global forgetting patterns and local consistency checks along a single teacher trajectory can separate clean from noisy samples reliably enough to guide reweighting and auxiliary targets.

What would settle it

If distilled datasets produced by this method yield no accuracy gain over standard distillation baselines when trained on the same noisy data and evaluated on clean test sets, the central claim would not hold.

Figures

Figures reproduced from arXiv: 2606.29837 by Fan Zhang, Jiyang Li, Kaifeng Chen, Lechao Cheng, Shengeng Tang, Tuanrui Hui, Yantao Pan, Yaxiong Wang, Zhun Zhong.

Figure 1
Figure 1. Figure 1: The overall pipeline. Our proposed pipeline includes two main components: (1) during teacher trajectory training, we apply sample-specific weighting adjustments to modulate the influence of each sample based on its estimated reliability; and (2) in the subsequent distillation phase, we leverage a subset of high-confidence samples to impose additional constraints and regularization, thereby enhancing the qu… view at source ↗
Figure 2
Figure 2. Figure 2: Performance comparison between Diverse Sampling and Fixed Sampling un￾der various settings on CIFAR-10. Baseline refers to DATM without our method. Diverse Sampling and Fixed Sampling incorporate our Selective Guidance Reweighting into DATM, where Fixed Sampling uses fixed αmax values (0, 0.5, 1.0) for teacher trajectory training, while Diverse Sampling multi follows the strategy outlined in Section4.2. to… view at source ↗
read the original abstract

Dataset distillation (DD) condenses large corpora into compact, information-rich subsets for efficient training and reuse. However, under noisy supervision, DD risks condensing corrupted associations together with useful signals, degrading robustness. Conventional noisy-label remedies (sample selection, loss weighting, label correction) tightly couple noise estimation with model optimization, often require clean anchors, and can amplify confirmation bias-assumptions that are misaligned with DD's goal of compact, plug-and-play supervision. We therefore propose a trajectory-based DD framework that jointly suppresses noise and preserves transferable knowledge without relabeling or clean subsets. It comprises two complementary components: Selective Guidance Reweighting (SGR), which fuses global forgetting patterns (second-split forgetting) with local neighborhood consistency into a progressive reweighting scheme that prioritizes clean supervision along the teacher trajectory; and Teacher-Inspired Auxiliary Targets (TIAT), which inject auxiliary residual guidance distilled from intermediate teacher dynamics to reinforce informative signals while remaining internally consistent. Together, SGR and TIAT produce distilled datasets with cleaner and richer representations under noisy supervision. The framework is robust, label-preserving, computationally lightweight, and broadly applicable, yielding consistent gains over state-of-the-art DD baselines across symmetric, asymmetric, and real-world noise.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper proposes Robust Trajectory Distillation, a framework for dataset distillation under noisy supervision. It introduces Selective Guidance Reweighting (SGR), which fuses second-split forgetting patterns with neighborhood consistency into a progressive reweighting scheme along the teacher trajectory, and Teacher-Inspired Auxiliary Targets (TIAT), which injects auxiliary residual guidance from intermediate teacher dynamics. The method claims to suppress noise while preserving transferable knowledge without relabeling or clean anchors, producing cleaner distilled datasets and yielding consistent gains over state-of-the-art DD baselines across symmetric, asymmetric, and real-world noise.

Significance. If the empirical claims hold and the reweighting reliably isolates clean signals, the work addresses a misalignment between standard noisy-label techniques and the goals of dataset distillation, offering a lightweight, label-preserving approach applicable to real-world noisy data in computer vision. This could enable more robust plug-and-play supervision from condensed datasets.

major comments (2)
  1. [SGR description (method section)] The central claim that SGR's fusion of global second-split forgetting with local neighborhood consistency produces a reweighting scheme monotonic in cleanliness (prioritizing clean supervision) is load-bearing but unsupported by any derivation or bound. In asymmetric noise, forgetting trajectories for noisy labels can overlap with clean ones after the first split, and TIAT's residual guidance from the same corrupted teacher does not resolve this dependence; no section provides a formal argument or test isolating this separation.
  2. [Experimental results] Experiments report consistent gains over DD baselines, but without ablations that hold the teacher fixed while varying noise asymmetry or that measure correlation between the combined SGR score and ground-truth cleanliness, it is unclear whether gains stem from the claimed mechanism or from other factors; this directly affects the robustness claim under asymmetric and real-world noise.
minor comments (2)
  1. [Abstract] The abstract states the framework is 'computationally lightweight' without quantifying training overhead or memory relative to baselines; add a table or paragraph with these metrics.
  2. [Method] Notation for 'second-split forgetting' and 'neighborhood consistency' should be defined with explicit formulas or pseudocode in the method section for reproducibility.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on our manuscript. We address each major comment point by point below, providing clarifications on the empirical foundations of our approach and indicating where revisions will be made to strengthen the presentation.

read point-by-point responses

  1. Referee: [SGR description (method section)] The central claim that SGR's fusion of global second-split forgetting with local neighborhood consistency produces a reweighting scheme monotonic in cleanliness (prioritizing clean supervision) is load-bearing but unsupported by any derivation or bound. In asymmetric noise, forgetting trajectories for noisy labels can overlap with clean ones after the first split, and TIAT's residual guidance from the same corrupted teacher does not resolve this dependence; no section provides a formal argument or test isolating this separation.

    Authors: We acknowledge that the manuscript does not include a formal derivation or theoretical bound establishing that the SGR reweighting is strictly monotonic in label cleanliness. The design of SGR is motivated by empirical patterns observed in forgetting trajectories and neighborhood consistency under noise, as described in the method section, rather than a closed-form proof. Deriving such a bound is non-trivial given the dependence on teacher trajectory dynamics and would require assumptions that may not hold across all noise regimes; we view this as beyond the current scope. We will revise the method section to explicitly note the empirical motivation, potential overlaps in asymmetric noise, and the reliance on experimental validation rather than theoretical guarantees. revision: partial

  2. Referee: [Experimental results] Experiments report consistent gains over DD baselines, but without ablations that hold the teacher fixed while varying noise asymmetry or that measure correlation between the combined SGR score and ground-truth cleanliness, it is unclear whether gains stem from the claimed mechanism or from other factors; this directly affects the robustness claim under asymmetric and real-world noise.

    Authors: The reported experiments already evaluate performance across symmetric, asymmetric, and real-world noise settings, with the teacher trained on the corresponding noisy data. However, we agree that an ablation fixing the teacher while varying noise asymmetry would more directly isolate SGR's contribution. Similarly, while we have not reported Pearson or Spearman correlations between SGR scores and ground-truth cleanliness (as real-world settings lack such labels), this can be computed on the synthetic noise benchmarks. We will add these targeted ablations and correlation analyses to the experimental section in the revision to better substantiate the mechanism. revision: yes

Circularity Check

0 steps flagged

No circularity; derivation uses external training observables

full rationale

The framework defines SGR via second-split forgetting and neighborhood consistency, and TIAT via residual teacher dynamics; both are computed from observable training trajectories rather than fitted to or defined by the final distilled dataset quality. No equations reduce the reweighting or targets to the target result by construction, no self-citation chain is load-bearing for the central claim, and no ansatz or uniqueness theorem is smuggled in. The approach remains falsifiable against external noise benchmarks without internal redefinition.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

Abstract-only review; full paper would be required to identify concrete free parameters in the reweighting formulas or target construction. The method rests on domain assumptions about the reliability of forgetting patterns and teacher dynamics as noise indicators.

axioms (2)
  • domain assumption Global forgetting patterns (second-split forgetting) fused with local neighborhood consistency can prioritize clean supervision along the teacher trajectory without clean anchors.
    Core premise of Selective Guidance Reweighting stated in abstract.
  • domain assumption Auxiliary residual guidance distilled from intermediate teacher dynamics reinforces informative signals while remaining internally consistent.
    Core premise of Teacher-Inspired Auxiliary Targets stated in abstract.

pith-pipeline@v0.9.1-grok · 5775 in / 1382 out tokens · 45785 ms · 2026-06-30T06:51:36.946713+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

51 extracted references · 25 canonical work pages · 3 internal anchors

  1. [1]

    In: International Conference on Machine Learning

    Bahri, D., Jiang, H., Gupta, M.: Deep k-nn for noisy labels. In: International Conference on Machine Learning. pp. 540–550. PMLR (2020)

    2020

  2. [2]

    Cazenavette, G., Wang, T., Torralba, A., Efros, A.A., Zhu, J.Y.: Dataset distillation by matching training trajectories (2022),https://arxiv.org/ abs/2203.11932

    work page arXiv 2022

  3. [3]

    org/abs/2310.06982

    Chen, X., Yang, Y., Wang, Z., Mirzasoleiman, B.: Data distillation can be like vodka: Distilling more times for better quality (2023),https://arxiv. org/abs/2310.06982

    work page arXiv 2023

  4. [4]

    arXiv preprint arXiv:2411.11924 (2024)

    Cheng, L., Chen, K., Li, J., Tang, S., Zhang, S., Wang, M.: Dataset distillers are good label denoisers in the wild. arXiv preprint arXiv:2411.11924 (2024)

    work page arXiv 2024

  5. [5]

    Cui, J., Wang, R., Si, S., Hsieh, C.J.: Dc-bench: Dataset condensation benchmark (2022),https://arxiv.org/abs/2207.09639

    work page arXiv 2022

  6. [6]

    Deng, W., Li, W., Ding, T., Wang, L., Zhang, H., Huang, K., Huo, J., Gao, Y.: Exploiting inter-sample and inter-feature relations in dataset distillation (2024),https://arxiv.org/abs/2404.00563

    work page arXiv 2024

  7. [7]

    arXiv preprint arXiv:2408.14358 (2024)

    Di Salvo, F., Doerrich, S., Rieger, I., Ledig, C.: An embedding is worth a thousand noisy labels. arXiv preprint arXiv:2408.14358 (2024)

    work page arXiv 2024

  8. [8]

    Du, J., Jiang, Y., Tan, V.Y.F., Zhou, J.T., Li, H.: Minimizing the accu- mulated trajectory error to improve dataset distillation (2023),https: //arxiv.org/abs/2211.11004

    work page arXiv 2023

  9. [9]

    In: ICLR 2024-The Twelfth International Conference on Learning Representations, Messe Wien Exhibition and Congress Center, Vienna, Austria, May 7-11t, 2024 (2024)

    Englesson, E., Azizpour, H.: Robust classification via regression for learning with noisy labels. In: ICLR 2024-The Twelfth International Conference on Learning Representations, Messe Wien Exhibition and Congress Center, Vienna, Austria, May 7-11t, 2024 (2024)

    2024

  10. [10]

    IEEE Transactions on Neural Networks and Learning Systems35(11), 16036–16048 (2023)

    Fang, C., Cheng, L., Mao, Y., Zhang, D., Fang, Y., Li, G., Qi, H., Jiao, L.: Separating noisy samples from tail classes for long-tailed image classifica- tion with label noise. IEEE Transactions on Neural Networks and Learning Systems35(11), 16036–16048 (2023)

    2023

  11. [11]

    IEEE Transactions on Medical Imaging42(6), 1720– 1734 (2023)

    Fang, C., Wang, Q., Cheng, L., Gao, Z., Pan, C., Cao, Z., Zheng, Z., Zhang, D.: Reliable mutual distillation for medical image segmentation under im- perfect annotations. IEEE Transactions on Medical Imaging42(6), 1720– 1734 (2023)

    2023

  12. [12]

    Guo, Z., Wang, K., Cazenavette, G., Li, H., Zhang, K., You, Y.: Towards lossless dataset distillation via difficulty-aligned trajectory matching (2024), https://arxiv.org/abs/2310.05773

    work page arXiv 2024

  13. [13]

    Advances in neural information processing systems31(2018)

    Han, B., Yao, Q., Yu, X., Niu, G., Xu, M., Hu, W., Tsang, I., Sugiyama, M.: Co-teaching: Robust training of deep neural networks with extremely noisy labels. Advances in neural information processing systems31(2018)

    2018

  14. [14]

    He, Y., Xiao, L., Zhou, J.T., Tsang, I.: Multisize dataset condensation (2024),https://arxiv.org/abs/2403.06075

    work page arXiv 2024

  15. [15]

    2022 ieee

    Iscen, A., Valmadre, J., Arnab, A., Schmid, C.: Learning with neighbor consistency for noisy labels. 2022 ieee. In: CVF Conference on Computer Vision and Pattern Recognition (CVPR). pp. 4662–4671 (2022) Robust Trajectory Distillation 17

    2022

  16. [16]

    In: International conference on machine learning

    Jiang, L., Zhou, Z., Leung, T., Li, L.J., Fei-Fei, L.: Mentornet: Learning data-driven curriculum for very deep neural networks on corrupted labels. In: International conference on machine learning. pp. 2304–2313. PMLR (2018)

    2018

  17. [17]

    Krizhevsky, A.: Learning multiple layers of features from tiny images (2009), https://api.semanticscholar.org/CorpusID:18268744

    2009

  18. [18]

    CS 231N7(7), 3 (2015)

    Le, Y., Yang, X.: Tiny imagenet visual recognition challenge. CS 231N7(7), 3 (2015)

    2015

  19. [19]

    Lee, Y., Chung, H.W.: Selmatch: Effectively scaling up dataset distillation via selection-based initialization and partial updates by trajectory matching (2024),https://arxiv.org/abs/2406.18561

    work page arXiv 2024

  20. [20]

    In: European Conference on Computer Vision

    Li, J., Li, G., Liu, F., Yu, Y.: Neighborhood collective estimation for noisy label identification and correction. In: European Conference on Computer Vision. pp. 128–145. Springer (2022)

    2022

  21. [21]

    arXiv preprint arXiv:2002.07394 , year=

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

    work page arXiv 2002

  22. [22]

    WebVision Database: Visual Learning and Understanding from Web Data

    Li, W., Wang, L., Li, W., Agustsson, E., Gool, L.V.: Webvision database: Visual learning and understanding from web data (2017),https://arxiv. org/abs/1708.02862

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  23. [23]

    Advances in neural information processing systems33, 20331–20342 (2020)

    Liu, S., Niles-Weed, J., Razavian, N., Fernandez-Granda, C.: Early-learning regularization prevents memorization of noisy labels. Advances in neural information processing systems33, 20331–20342 (2020)

    2020

  24. [24]

    In: International conference on machine learning

    Liu, Y., Guo, H.: Peer loss functions: Learning from noisy labels without knowing noise rates. In: International conference on machine learning. pp. 6226–6236. PMLR (2020)

    2020

  25. [25]

    Loo, N., Hasani, R., Lechner, M., Rus, D.: Dataset distillation with convex- ified implicit gradients (2023),https://arxiv.org/abs/2302.06755

    work page arXiv 2023

  26. [26]

    arXiv preprint arXiv:1905.10045 (2019)

    Lyu, Y., Tsang, I.W.: Curriculum loss: Robust learning and generalization against label corruption. arXiv preprint arXiv:1905.10045 (2019)

    work page arXiv 1905

  27. [27]

    Advances in Neural Information Processing Systems 35, 30044–30057 (2022)

    Maini, P., Garg, S., Lipton, Z., Kolter, J.Z.: Characterizing datapoints via second-split forgetting. Advances in Neural Information Processing Systems 35, 30044–30057 (2022)

    2022

  28. [28]

    when to update

    Malach, E., Shalev-Shwartz, S.: Decoupling" when to update" from" how to update". Advances in neural information processing systems30(2017)

    2017

  29. [29]

    CoRRabs/2107.13034(2021),https:// arxiv.org/abs/2107.13034

    Nguyen, T., Novak, R., Xiao, L., Lee, J.: Dataset distillation with infinitely wide convolutional networks. CoRRabs/2107.13034(2021),https:// arxiv.org/abs/2107.13034

    work page arXiv 2021

  30. [30]

    Training Deep Neural Networks on Noisy Labels with Bootstrapping

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

    work page internal anchor Pith review Pith/arXiv arXiv 2014

  31. [31]

    Sachdeva, N., McAuley, J.: Data distillation: A survey (2023),https:// arxiv.org/abs/2301.04272

    work page arXiv 2023

  32. [32]

    Advances in neural information processing systems32(2019) 18 Kaifeng Chen et al

    Shu, J., Xie, Q., Yi, L., Zhao, Q., Zhou, S., Xu, Z., Meng, D.: Meta-weight- net: Learning an explicit mapping for sample weighting. Advances in neural information processing systems32(2019) 18 Kaifeng Chen et al

    2019

  33. [33]

    In: International conference on machine learning

    Song,H.,Kim,M.,Lee,J.G.:Selfie:Refurbishinguncleansamplesforrobust deep learning. In: International conference on machine learning. pp. 5907–

  34. [34]

    IEEE transactions on neural networks and learning systems34(11), 8135–8153 (2022)

    Song, H., Kim, M., Park, D., Shin, Y., Lee, J.G.: Learning from noisy labels with deep neural networks: A survey. IEEE transactions on neural networks and learning systems34(11), 8135–8153 (2022)

    2022

  35. [35]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) (2024)

    Sun, P., Shi, B., Yu, D., Lin, T.: On the diversity and realism of dis- tilled dataset: An efficient dataset distillation paradigm. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) (2024)

    2024

  36. [36]

    In: Pro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Tu, Y., Zhang, B., Li, Y., Liu, L., Li, J., Wang, Y., Wang, C., Zhao, C.R.: Learning from noisy labels with decoupled meta label purifier. In: Pro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 19934–19943 (2023)

    2023

  37. [37]

    Wang, S., Yang, Y., Liu, Z., Sun, C., Hu, X., He, C., Zhang, L.: Dataset dis- tillation with neural characteristic function: A minmax perspective (2025), https://arxiv.org/abs/2502.20653

    work page arXiv 2025

  38. [38]

    In: 2018 IEEE 30th international conference on tools with artificial intelligence (ICTAI)

    Wang, T., Huan, J., Li, B.: Data dropout: Optimizing training data for convolutional neural networks. In: 2018 IEEE 30th international conference on tools with artificial intelligence (ICTAI). pp. 39–46. IEEE (2018)

    2018

  39. [39]

    Dataset Distillation

    Wang, T., Zhu, J., Torralba, A., Efros, A.A.: Dataset distillation. CoRR abs/1811.10959(2018),http://arxiv.org/abs/1811.10959

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  40. [40]

    In: European Conference on Computer Vision

    Wang, Y., Cheng, L., Duan, M., Wang, Y., Feng, Z., Kong, S.: Improv- ing knowledge distillation via regularizing feature direction and norm. In: European Conference on Computer Vision. pp. 20–37. Springer (2024)

    2024

  41. [41]

    Wei, J., Zhu, Z., Cheng, H., Liu, T., Niu, G., Liu, Y.: Learning with noisy labelsrevisited:Astudyusingreal-worldhumanannotations.arXivpreprint arXiv:2110.12088 (2021)

    work page arXiv 2021

  42. [42]

    Zhang, H., Li, S., Lin, F., Wang, W., Qian, Z., and Ge, S

    Zhang, H., Li, S., Lin, F., Wang, W., Qian, Z., Ge, S.: Dance: Dual-view distribution alignment for dataset condensation (2024),https://arxiv. org/abs/2406.01063

    work page arXiv 2024

  43. [43]

    Zhang, H., Li, S., Wang, P., Zeng, D., Ge, S.: M3d: Dataset condensation by minimizing maximum mean discrepancy (2024),https://arxiv.org/ abs/2312.15927

    work page arXiv 2024

  44. [44]

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

    Zhang, T., Xue, M., Zhang, J., Zhang, H., Wang, Y., Cheng, L., Song, J., Song, M.: Generalization matters: Loss minima flattening via parameter hybridization for efficient online knowledge distillation. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 20176–20185 (2023)

    2023

  45. [45]

    Advances in neural information process- ing systems31(2018)

    Zhang, Z., Sabuncu, M.: Generalized cross entropy loss for training deep neural networks with noisy labels. Advances in neural information process- ing systems31(2018)

    2018

  46. [46]

    Zhao, B., Bilen, H.: Dataset condensation with distribution matching (2022),https://arxiv.org/abs/2110.04181

    work page arXiv 2022

  47. [47]

    Zhao, B., Mopuri, K.R., Bilen, H.: Dataset condensation with gradient matching (2021),https://arxiv.org/abs/2006.05929 Robust Trajectory Distillation 19

    work page arXiv 2021

  48. [48]

    In: International Conference on Learning Representations (2021)

    Zhou, T., Wang, S., Bilmes, J.: Robust curriculum learning: from clean label detection to noisy label self-correction. In: International Conference on Learning Representations (2021)

    2021

  49. [49]

    Zhou, Y., Nezhadarya, E., Ba, J.: Dataset distillation using neural feature regression (2022),https://arxiv.org/abs/2206.00719

    work page arXiv 2022

  50. [50]

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

    Zhou, Y., Li, X., Liu, F., Wei, Q., Chen, X., Yu, L., Xie, C., Lungren, M.P., Xing, L.: L2b: Learning to bootstrap robust models for combating label noise. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 23523–23533 (2024)

    2024

  51. [51]

    In: International conference on machine learning

    Zhu, Z., Dong, Z., Liu, Y.: Detecting corrupted labels without training a model to predict. In: International conference on machine learning. pp. 27412–27427. PMLR (2022)

    2022