Pith. sign in

REVIEW 2 major objections 5 minor 71 references

Noisy ranks are treated as probabilistic associations with several neighboring ranks, and an embedding is learned that respects those associations.

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

Stochastic Order Learning associates each instance with multiple plausible ranks and trains embeddings via complementary discriminative and stochastic-order losses that remain robust to ordinal label noise.

T0 review reviewed 2026-07-10 challenge →

load-bearing objection Clean, usable extension of geometric order learning to noisy ordinal labels; the stochastic reformulation is real and the empirics are broad enough to trust the gains. the 2 major comments →

arxiv 2607.08103 v1 pith:Z6V5UZQJ submitted 2026-07-09 cs.LG

Stochastic Order Learning: An Approach to Rank Estimation Using Noisy Data

classification cs.LG
keywords rank estimationordinal regressionlabel noiseorder learningstochastic orderingembedding spacenoisy supervision
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The reading

Rank labels in age estimation, aesthetics, medical scoring and similar tasks are rarely exact; nearby ranks are far more likely than distant ones, so ordinary classification or regression that treats every wrong label the same is poorly matched to the problem. The paper reframes the task as stochastic ordering: each training example is allowed to relate, with decaying probability, to a whole neighborhood of ranks rather than to a single correct one. From that view it builds Stochastic Order Learning, which places instances and rank centroids in a shared embedding so that expected squared distances obey a monotonicity condition. Two losses cooperate to enforce the condition: a discriminative loss that attracts each point toward plausible centroids and repels it from implausible ones, and a stochastic order loss that keeps relative orderings between pairs consistent with the noise model. Optional outlier detection and mild relabeling further clean extreme errors. Experiments on facial age, aesthetic scores, bone-age radiographs and translation quality scores show that the resulting estimator remains accurate under Gaussian, Laplacian, uniform and real-world annotation noise.

Core claim

When ordinal labels are noisy, each instance should be treated as stochastically associated with multiple neighboring ranks; an embedding that minimizes the expected squared distance to the corresponding rank centroids, together with pairwise probabilistic order constraints, yields rank estimates that stay reliable across noise types and severities.

What carries the argument

Stochastic dissimilarity Dh(x,r) = sum_s ps ||h(x) - mu_{r+s}||^2, whose minimization under a discrete-Gaussian noise model produces the monotonicity constraint that the discriminative and stochastic-order losses jointly enforce.

Load-bearing premise

The method derives its losses and centroid rule from a discrete Gaussian model of label errors; if real annotation noise is strongly asymmetric or multi-modal, those derived objectives may no longer match the data.

What would settle it

On a large ordinal dataset whose annotation errors are known to be strongly skewed or multi-modal, replace the Gaussian ps with the true error frequencies; if SOL then loses its advantage over ordinary order-learning and noise-robust regression baselines, the Gaussian modeling assumption is essential rather than merely convenient.

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

If this is right

  • Noise-robust rank estimators can be obtained without requiring clean reference labels or explicit noise-rate estimation.
  • The same stochastic-order losses apply unchanged to vision, medical imaging and text regression once an appropriate encoder is chosen.
  • Outlier detection and mild relabeling become natural by-products of the learned embedding rather than separate heuristics.
  • Inference reduces to finding the rank whose expected squared distance is smallest, avoiding expensive nearest-neighbor search at test time.

Where Pith is reading between the lines

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

  • If annotator-specific or input-dependent noise distributions can be estimated on the fly, the same Dh formulation could absorb them without redesigning the losses.
  • The monotonicity constraint may transfer to continuous ordinal regression by replacing discrete centroids with a smooth rank manifold.
  • Because the method already models structured uncertainty, it is a natural candidate for active learning of the most ambiguous ranks.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 5 minor

Summary. The paper reformulates rank estimation under noisy ordinal labels as a stochastic ordering problem: each instance is associated with a distribution over plausible ranks rather than a single deterministic label. It introduces Stochastic Order Learning (SOL), which learns an embedding via a discriminative loss that attracts/repels instances relative to rank centroids under a stochastic dissimilarity Dh (Eqs. 3–10) and a stochastic order loss that enforces probabilistic pairwise ordering relations (Eqs. 12–17). Centroids are updated in closed form (Eq. 18); an optional outlier detection and relabeling step (Eqs. 19–21) is also provided. Inference selects the rank minimizing expected distance (Eq. 6). Experiments on MORPH II, CLAP2015, AADB, RSNA, and WMT2020 under Gaussian, Laplacian, uniform, skewed, and real-world noise show consistent gains over noise-robust classification/regression and order-learning baselines.

Significance. If the results hold, SOL supplies a clean, generalizable framework for ordinal prediction under structured label noise—an important practical setting that standard classification noise methods and clean-label order learning both handle poorly. Strengths include: (i) explicit derivations of the monotonicity constraint (Appendix A) and closed-form centroid rule (Appendix B); (ii) complementary losses that are not tautological restatements of the evaluation metric; (iii) broad empirical coverage across five datasets, four synthetic noise families, multiple κ levels, real-world WMT2020 noise, ablations of each loss, multi-seed stability, and alternative relabeling schemes; and (iv) released source code. The free parameters (T, τ, γ, β, σ_test) are documented and analyzed. The contribution is incremental relative to prior order learning (GOL, MWR) but the stochastic reformulation and the resulting robustness gains are concrete and useful.

major comments (2)
  1. Main-result Tables 1–4 report only point estimates of MAE/CS with no standard deviations or multi-seed statistics. Appendix D.12 later supplies five-seed means±std on MORPH II only. Because the central claim is consistent outperformance under diverse noise, the main tables should either include error bars (or multi-seed means) for all five datasets or explicitly state that the Appendix D.12 stability analysis is representative; otherwise the magnitude of the reported gains (often 0.02–0.1 MAE) cannot be assessed for statistical reliability.
  2. Section 3.1 and Eq. (2) introduce a discrete Gaussian for the noise model, and the desideratum, losses, and centroid rule are written with general {ps}. The paper correctly stress-tests Laplacian/uniform/skewed/input-dependent noise and real WMT2020 annotations (Tables 1–5, Appendix D.2). However, the main text still presents the Gaussian as the default generative model without a short, explicit statement that the algorithmic objects remain well-defined for any unimodal discrete {ps}. Adding one clarifying sentence (and, if space permits, a one-line general-ps derivation note) would remove residual ambiguity about how load-bearing the parametric form is.
minor comments (5)
  1. Figure 1 caption and surrounding text correctly contrast nominal vs. ordinal noise severity; a brief pointer to the later quantitative tables would help readers who skip the figure.
  2. Hyperparameter defaults (Table 9) and sensitivity analyses (Appendix D.1) are thorough; moving a one-sentence summary of recommended defaults into the main Implementation subsection would improve reproducibility for readers who do not open the appendix.
  3. Notation: the same symbol e is used both for the random noise variable and for individual errors ex; a minor typographic distinction would avoid momentary confusion in Section 3.1.
  4. Impact Statement correctly flags demographic bias risk on facial datasets; a parallel one-sentence note for the medical (RSNA) setting would be consistent.
  5. A few typographical slips remain (e.g., “order-identitiy” in Related Work; occasional missing spaces around citations). A final proofreading pass is warranted.

Circularity Check

0 steps flagged

No significant circularity: SOL's losses and centroid rule are derived from an explicit desideratum under a stated noise model, then evaluated on external benchmarks against independent baselines.

full rationale

The paper proposes a stochastic reformulation of noisy rank estimation (Eqs. 1–4), derives a sufficient monotonicity condition (Eq. 5, Appendix A), constructs a discriminative surrogate loss (Eqs. 8–10) and a probabilistic pairwise order loss (Eqs. 12–17) that encourage the desideratum, and obtains the centroid update by direct minimization of the expected squared distance (Eq. 18, Appendix B). Inference simply re-uses the same Dh (Eq. 6). None of these steps is self-definitional: the desideratum is an explicit design goal, not a restatement of the evaluation metrics (MAE/CS/PCC/SRCC). The discrete-Gaussian form of ps is an assumption, not a fitted quantity later re-presented as a prediction; the same algebraic forms hold for any discrete {ps}, and the paper empirically stress-tests Laplacian, uniform, skewed and input-dependent noise plus real WMT2020 annotations. Self-citations to prior order-learning papers (Lim et al., Lee et al.) supply only the clean-label baseline architecture; the stochastic losses, centroid rule and outlier module are new and are not justified by any uniqueness theorem or load-bearing self-citation. No parameter is fitted on a subset and then called a prediction of a related quantity. The method is therefore self-contained against external public benchmarks and independent baselines; circularity score is zero.

Axiom & Free-Parameter Ledger

5 free parameters · 3 axioms · 2 invented entities

The central claim rests on a parametric noise model, a small set of hand-chosen hyper-parameters that control attraction range and outlier aggressiveness, and the invented stochastic dissimilarity and order losses. These are the quantities that must be accepted for the reported robustness to hold.

free parameters (5)
  • T (summation range in discriminative loss)
    Controls how many neighboring ranks participate in attraction/repulsion; default T=1 chosen by ablation on CLAP2015.
  • τ (order-case threshold)
    Defines the three probabilistic order relations ≺, ≈, ≻; set to 3 or 5 by validation.
  • γ (margin in order losses)
    Fixed at 0.25 across all data sets; not derived.
  • β (outlier detection fraction)
    Controls precision of the relabeling step; set to 0.85–0.9 by hand.
  • σ_test (fixed noise scale at inference)
    Used to compute ps when true noise level is unknown; chosen by sensitivity study, not estimated from data.
axioms (3)
  • domain assumption Label noise e follows a discrete Gaussian (symmetric, unimodal) distribution whose probabilities weight both Dh and the pairwise order probabilities.
    Stated in Section 3.1 Eq. (2); the entire loss derivation proceeds from this form.
  • standard math The monotonicity constraint d(hx, μ_rx+s) ≤ d(hx, μ_rx+t) for |s|≤|t| is a sufficient condition for the stochastic-dissimilarity desideratum.
    Proved in Appendix A under the unimodal-symmetric assumption on ps.
  • domain assumption Centroids that minimize the expected squared distance under the same noise model are the optimal representatives of each rank.
    Derived in Appendix B and used for the closed-form update (18).
invented entities (2)
  • Stochastic dissimilarity Dh(x,r) no independent evidence
    purpose: Replaces deterministic distance-to-rank with an expectation over neighboring centroids weighted by the noise probabilities.
    Defined in Eq. (3); the entire training objective is built around minimizing or ordering these quantities.
  • Discriminative loss ℓ_disc and stochastic order loss ℓ_order no independent evidence
    purpose: Two complementary surrogates that enforce the desideratum and the probabilistic pairwise order relations, respectively.
    Introduced in Section 3.2; ablation shows both are required for good performance.

reviewed 2026-07-10 · how reviews work

0 comments
Cite this review

Pith. "Pith review of Stochastic Order Learning: An Approach to Rank Estimation Using Noisy Data." pith.science (2026). https://pith.science/paper/Z6V5UZQJ

@misc{pith2026260708103,
  author       = {Pith},
  title        = {Pith review of: Stochastic Order Learning: An Approach to Rank Estimation Using Noisy Data},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/Z6V5UZQJ}},
  note         = {Machine review of arXiv:2607.08103}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Rank estimation under label noise poses a fundamental challenge, as ordinal annotations often exhibit structured uncertainty rather than simple label corruption. In this paper, we reformulate rank estimation with noisy ordinal labels as a stochastic ordering problem, in which each instance is inherently associated with multiple plausible ranks instead of a single deterministic label. Based on this view, we propose stochastic order learning (SOL), a learning framework that captures ordinal label uncertainty and learns an embedding space through two complementary objectives: a discriminative loss that structures instance--centroid interactions and a stochastic order loss that enforces probabilistic ordering relations between instances. Extensive experiments across diverse datasets demonstrate that SOL enables reliable rank estimation under various types and levels of label noise. The source code is available at https://github.com/cwlee00/SOL.

Figures

Figures reproduced from arXiv: 2607.08103 by Chaewon Lee, Chang-Su Kim, Seon-Ho Lee.

Figure 1
Figure 1. Figure 1: Nominal data in classification versus ordered data in rank estimation. Unlike classification, in rank estimation, certain errors are severer than others. as a marginal perturbation to otherwise deterministic super￾vision. Rather, rank estimation under label noise must be reformulated as a stochastic ordering problem, in which in￾stance–rank relationships are inherently probabilistic rather than fixed. In t… view at source ↗
Figure 2
Figure 2. Figure 2: Overview of the proposed SOL algorithm closer ordinal labels more frequently. Wang et al. (2022a) showed that standard regularization schemes are ineffective under label noise and proposed a noise-robust text regres￾sion method that mitigates noise by discarding or repairing detected noisy samples. More recently, Kim et al. (2024) in￾troduced a contrastive fragmentation strategy that partitions the label s… view at source ↗
Figure 3
Figure 3. Figure 3: Illustration of the monotonicity constraint and the training losses for constructing a SOL embedding space µr¯x . Specifically, x is associated with µrx−s with probabil￾ity ps in (2). Note that, due to the symmetry ps = p−s, x is also associated with µrx+s with ps. Thus, in the embed￾ding space, the mean squared distance P s psd 2 (hx, µrx+s) should be minimized, where d denotes the Euclidean dis￾tance and… view at source ↗
Figure 4
Figure 4. Figure 4: (a) Success and (b) failure cases of age estimation results on the MORPH II dataset. Under each image, we compare the estimated ages of SPR (Wang et al., 2022b), GOL (Lee et al., 2022), and the proposed SOL and specify the corresponding errors inside the parentheses. rank labels, which may rather increase the label errors af￾ter relabeling. We repeat the encoder fine-tuning and the outlier detection and re… view at source ↗
Figure 5
Figure 5. Figure 5: (a) Success and (b) failure cases of the label refinement on the RSNA training dataset. Under each image, the noisy, refined, and true ranks are specified: noisy → refined (true). by 2.5% in terms of MAE and improves the CS over the best competing scores by 1.1 percentage points. Even at the lowest κ = 0.2, SOL reduces the MAE by 2.6% and improves the CS by 0.3 percentage points over GOL. Medical assessmen… view at source ↗
Figure 6
Figure 6. Figure 6: Comparison of the CS curves according to tolerance values on the RSNA dataset (Gaussian label noise). The legend of each graph includes the AUC score for the corresponding algorithm. 14 [PITH_FULL_IMAGE:figures/full_fig_p014_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Network structure of the encoder h. C.6. Hyperparameter Settings For WMT2020, we train the network for 20 epochs. For all the other datasets, we train the network for 100 epochs [PITH_FULL_IMAGE:figures/full_fig_p015_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: MAE according to σtest on CLAP2015. 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0.0 0.8 𝜎test = 0.5 𝜎test = 1.0 𝜎test = 2.0 𝜎test = 3.0 𝜎test = 4.0 𝜎test = 5.0 𝜎test = 6.0 Probability [PITH_FULL_IMAGE:figures/full_fig_p017_8.png] view at source ↗
Figure 10
Figure 10. Figure 10: Comparison of the numbers of instances at each noise level before and after the label refinement on the MORPH II dataset. D.5. Outliers in the WMT2020 Dataset We provide a qualitative analysis of outlier cases detected by SOL on the real-noise WMT2020 translation-quality dataset. Unlike synthetic noise, discrepancies in WMT2020 originate from genuine human variability, including strong penalties applied t… view at source ↗
Figure 11
Figure 11. Figure 11: MAE results according to the noise ratio κ on CLAP2015. 𝜅 = 0 𝜅 = 0.2 𝜅 = 0.4 3~7 8~12 13~17 18~22 23~27 28~32 33~37 38~42 43~47 48~52 53~57 58~62 63~67 68~72 73~77 78~82 83~87 True age [PITH_FULL_IMAGE:figures/full_fig_p023_11.png] view at source ↗
Figure 13
Figure 13. Figure 13: t-SNE visualization of the embedding spaces for MORPH II at different noise ratios κ. 𝜅 = 0 𝜅 = 0.2 𝜅 = 0.4 0.01~0.1 0.11~0.2 0.21~0.3 0.31~0.4 0.41~0.5 0.51~0.6 0.61~0.7 0.71~0.8 0.81~0.9 True score 0.91~1.0 [PITH_FULL_IMAGE:figures/full_fig_p025_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: t-SNE visualization of the embedding spaces for AADB at different noise ratios κ. 𝜅 = 0 𝜅 = 0.1 𝜅 = 0.2 0~19 20~39 40~59 60~79 80~99 100~119 120~139 140~159 160~179 180~199 200~219 True bone age 220~239 [PITH_FULL_IMAGE:figures/full_fig_p025_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: t-SNE visualization of the embedding spaces for RSNA at different noise ratios κ. 25 [PITH_FULL_IMAGE:figures/full_fig_p025_15.png] view at source ↗
Figure 16
Figure 16. Figure 16: (a) Success and (b) failure cases of age estimation results on the CLAP2015 dataset. Under each image, the estimated ages are specified with the ground-truth in parentheses. 26 [PITH_FULL_IMAGE:figures/full_fig_p026_16.png] view at source ↗
Figure 17
Figure 17. Figure 17: (a) Success and (b) failure cases of aesthetic score estimation results on the AADB dataset. Under each image, the estimated scores are specified with the ground-truth in parentheses. 27 [PITH_FULL_IMAGE:figures/full_fig_p027_17.png] view at source ↗
Figure 18
Figure 18. Figure 18: (a) Success and (b) failure cases of bone age assessment results on the RSNA dataset. Under each image, the estimated ages (in months) are specified with the ground-truth in parentheses. 28 [PITH_FULL_IMAGE:figures/full_fig_p028_18.png] view at source ↗
Figure 19
Figure 19. Figure 19: (a) Success and (b) failure cases of the label refinement on the MORPH II dataset. Under each image, the noisy, refined, and true ranks are specified: noisy → refined (true). 29 [PITH_FULL_IMAGE:figures/full_fig_p029_19.png] view at source ↗
Figure 20
Figure 20. Figure 20: (a) Success and (b) failure cases of the label refinement on the CLAP2015 dataset. Under each image, the noisy, refined, and true ranks are specified: noisy → refined (true). 30 [PITH_FULL_IMAGE:figures/full_fig_p030_20.png] view at source ↗
Figure 21
Figure 21. Figure 21: (a) Success and (b) failure cases of the label refinement on the AADB dataset. Under each image, the noisy, refined, and true ranks are specified: noisy → refined (true). 31 [PITH_FULL_IMAGE:figures/full_fig_p031_21.png] view at source ↗

discussion (0)

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

Reference graph

Works this paper leans on

71 extracted references · 71 canonical work pages · 1 internal anchor

  1. [1]

    Scaling Learning Algorithms Towards

    Bengio, Yoshua and LeCun, Yann , booktitle =. Scaling Learning Algorithms Towards

  2. [2]

    and Osindero, Simon and Teh, Yee Whye , journal =

    Hinton, Geoffrey E. and Osindero, Simon and Teh, Yee Whye , journal =. A Fast Learning Algorithm for Deep Belief Nets , volume =

  3. [3]

    2016 , publisher=

    Deep learning , author=. 2016 , publisher=

  4. [4]

    Visualizing data using t-

    Maaten, Laurens van der and Hinton, Geoffrey , journal=. Visualizing data using t-

  5. [5]

    CVPR , year=

    Human age estimation using bio-inspired features , author=. CVPR , year=

  6. [6]

    ICCV , year=

    Semi-supervised semantic segmentation under label noise via diverse learning groups , author=. ICCV , year=

  7. [7]

    Hwang, Seong-Hyeon and Kim, Minsu and Whang, Steven Euijong , booktitle=

  8. [8]

    Castells, Thibault and Weinzaepfel, Philippe and Revaud, Jerome , booktitle=

  9. [9]

    CVPR , year=

    Scalable penalized regression for noise detection in learning with noisy labels , author=. CVPR , year=

  10. [10]

    Very deep convolutional networks for large-scale image recognition , author=

  11. [11]

    Kingma and Jimmy L

    Diederik P. Kingma and Jimmy L. Ba , booktitle=ICLR, year=. Adam:

  12. [12]

    Deng, Jia and Dong, Wei and Socher, Richard and Li, Li-Jia and Li, Kai and Fei-Fei, Li , booktitle=CVPR, year=. Image

  13. [13]

    A constrained deep neural network for ordinal regression , author=

  14. [14]

    Soft labels for ordinal regression , author=

  15. [15]

    ECML , year=

    A simple approach to ordinal classification , author=. ECML , year=

  16. [16]

    2020 , booktitle=

    Robust deep ordinal regression under label noise , author=. 2020 , booktitle=

  17. [17]

    AAAI , year=

    Robust loss functions under label noise for deep neural networks , author=. AAAI , year=

  18. [18]

    Training deep neural-networks using a noise adaptation layer , author=

  19. [19]

    Masking:

    Han, Bo and Yao, Jiangchao and Niu, Gang and Zhou, Mingyuan and Tsang, Ivor and Zhang, Ya and Sugiyama, Masashi , booktitle=NIPS, year=. Masking:

  20. [20]

    Co-teaching:

    Han, Bo and Yao, Quanming and Yu, Xingrui and Niu, Gang and Xu, Miao and Hu, Weihua and Tsang, Ivor and Sugiyama, Masashi , booktitle=NIPS, year=. Co-teaching:

  21. [21]

    Jiang, Lu and Zhou, Zhengyuan and Leung, Thomas and Li, Li-Jia and Fei-Fei, Li , booktitle=ICLR, year=. Mentor

  22. [22]

    Photo aesthetics ranking network with attributes and content adaptation , author=

  23. [23]

    Order learning using partially ordered data via chainization , author=

  24. [24]

    Geometric order learning for rank estimation , author=

  25. [25]

    Unsupervised order learning , author=

  26. [26]

    Moving window regression:

    Shin, Nyeong-Ho and Lee, Seon-Ho and Kim, Chang-Su , booktitle=CVPR, year=. Moving window regression:

  27. [27]

    Deep Repulsive Clustering of Ordered Data Based on Order-Identitiy Decomposition , author=

  28. [28]

    Ordinal regression by extended binary classification , author=

  29. [29]

    Li, Wanhua and Lu, Jiwen and Feng, Jianjiang and Xu, Chunjing and Zhou, Jie and Tian, Qi , booktitle=CVPR, year=. Bridge

  30. [30]

    Learning Probabilistic Ordinal Embeddings for Uncertainty-Aware Regression , author=

  31. [31]

    Order learning and its application to age estimation , author=

  32. [32]

    Normalized loss functions for deep learning with noisy labels , author=

  33. [33]

    ICLR , year=

    Can gradient clipping mitigate label noise? , author=. ICLR , year=

  34. [34]

    NIPS , year =

    Active Negative Loss Functions for Learning with Noisy Labels , author =. NIPS , year =

  35. [35]

    Learning with noisy labels , author=

  36. [36]

    Mean-variance loss for deep age estimation from a face , author=

  37. [37]

    ICCV Workshops , year=

    Escalera, Sergio and Fabian, Junior and Pardo, Pablo and Bar\'. ICCV Workshops , year=

  38. [38]

    Ricanek, Karl and Tesafaye, Tamirat , booktitle=

  39. [39]

    Selfie: Refurbishing unclean samples for robust deep learning , author=

  40. [40]

    CVPR , year=

    Learning from noisy labels by regularized estimation of annotator confusion , author=. CVPR , year=

  41. [41]

    ICLR , year=

    Robust early-learning: Hindering the memorization of noisy labels , author=. ICLR , year=

  42. [42]

    ECCV , year=

    Learning with noisy class labels for instance segmentation , author=. ECCV , year=

  43. [43]

    Yao, Huaxiu and Wang, Yiping and Zhang, Linjun and Zou, James and Finn, Chelsea , booktitle=

  44. [44]

    mixup: Beyond empirical risk minimization , author=

  45. [45]

    Generalized cross entropy loss for training deep neural networks with noisy labels , author=

  46. [46]

    Ordinal hyperplanes ranker with cost sensitivities for age estimation , author=

  47. [47]

    CVPR , year=

    Deep regression forests for age estimation , author=. CVPR , year=

  48. [48]

    Sample selection via contrastive fragmentation for noisy label regression , author=

  49. [49]

    Proceedings of the 29th International Conference on Computational Linguistics , year=

    Noisy label regularisation for textual regression , author=. Proceedings of the 29th International Conference on Computational Linguistics , year=

  50. [50]

    Specia, Lucia and Blain, Fr \'e d \'e ric and Fomicheva, Marina and Fonseca, Erick and Chaudhary, Vishrav and Guzm \'a n, Francisco and Martins, Andr \'e F. T. Findings of the WMT 2020 Shared Task on Quality Estimation. Proceedings of the Fifth Conference on Machine Translation

  51. [51]

    ICLR , year=

    Improving Deep Regression with Ordinal Entropy , author=. ICLR , year=

  52. [52]

    ICML , year=

    Learning to rank using gradient descent , author=. ICML , year=

  53. [53]

    Proceedings of the 2008 International Conference on Web Search and Data Mining , year=

    Softrank: optimizing non-smooth rank metrics , author=. Proceedings of the 2008 International Conference on Web Search and Data Mining , year=

  54. [54]

    CVPR , year=

    Selective-supervised contrastive learning with noisy labels , author=. CVPR , year=

  55. [55]

    ICML , year=

    Peer loss functions: Learning from noisy labels without knowing noise rates , author=. ICML , year=

  56. [56]

    David Evans , booktitle=

    Mahsa Keramati and Lili Meng and R. David Evans , booktitle=

  57. [57]

    Pitawela, Dileepa and Carneiro, Gustavo and Chen, Hsiang-Ting , booktitle=

  58. [58]

    WACV , year=

    Distribution Highlighted Reference-based Label Distribution Learning for Facial Age Estimation , author=. WACV , year=

  59. [59]

    IJCAI , year=

    Age estimation using expectation of label distribution learning , author=. IJCAI , year=

  60. [60]

    CVPR , year=

    Unimodal-concentrated loss: Fully adaptive label distribution learning for ordinal regression , author=. CVPR , year=

  61. [61]

    Halabi, Safwan S and Prevedello, Luciano M and Kalpathy-Cramer, Jayashree and Mamonov, Artem B and Bilbily, Alexander and Cicero, Mark and Pan, Ian and Pereira, Lucas Ara. The. Radiology , volume=

  62. [62]

    Neurocomputing , volume=

    Unimodal regularized neuron stick-breaking for ordinal classification , author=. Neurocomputing , volume=

  63. [63]

    Neurocomputing , pages=

    Distributed robust support vector ordinal regression under label noise , author=. Neurocomputing , pages=

  64. [64]

    Curriculum Loss: Robust Learning and Generalization against Label Corruption

    Curriculum loss: Robust learning and generalization against label corruption , author=. arXiv preprint arXiv:1905.10045 , year=

  65. [65]

    Visualizing data using t-

    Van der Maaten, Laurens and Hinton, Geoffrey , journal=. Visualizing data using t-

  66. [66]

    IEEE Transactions on pattern analysis and machine intelligence , volume=

    Automatic age estimation based on facial aging patterns , author=. IEEE Transactions on pattern analysis and machine intelligence , volume=

  67. [67]

    TIP , volume=

    Deep label distribution learning with label ambiguity , author=. TIP , volume=

  68. [68]

    Journal of the Royal Statistical Society: Series B (Methodological) , volume=

    Regression models for ordinal data , author=. Journal of the Royal Statistical Society: Series B (Methodological) , volume=

  69. [69]

    Journal of the Royal Statistical Society: Series B (Methodological) , volume=

    Regression and ordered categorical variables , author=. Journal of the Royal Statistical Society: Series B (Methodological) , volume=

  70. [70]

    INFORMS Journal on Computing , volume=

    Isotonic separation , author=. INFORMS Journal on Computing , volume=

  71. [71]

    Rule learning with monotonicity constraints , author=

This paper was first reviewed by grok-4.5 on July 10, 2026.