Mixed Blessing: Class-Wise Embedding guided Instance-Dependent Partial Label Learning

Fuchao Yang; Hui Liu; Jianhong Cheng; Junhui Hou; Yongqiang Dong; Yuheng Jia

REVIEW 5 major objections 4 minor 43 references

Mixed Blessing: Class-Wise Embedding guided Instance-Dependent Partial Label Learning

T0 review · 5 major / 4 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read CEL claims that in instance-dependent partial label learning, per-class embeddings with two new losses turn the noisy labels' mixed blessing into a measurable accuracy gain, reaching the top of all six reported benchmarks.

desk verdict Solid empirical paper on instance-dependent PLL; gains look real, but the prototype self-training loop is underspecified and needs fixing before publication. read the letter →

arxiv 2412.05029 v1 pith:TCNI73FO submitted 2024-12-06 cs.LG

Fuchao Yang , Jianhong Cheng , Hui Liu , Yongqiang Dong , Yuheng Jia , Junhui Hou This is my paper · ORCID

classification cs.LG

keywords partiallabellearninginstance-dependentclass-wiseembeddingdisambiguationclassprototypesweaklysupervisedfine-grainedimageclassification

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 argues that instance-dependent partial label learning is a mixed blessing: the candidate labels are noisy, yet because they are generated from the image itself they carry genuine information about the sample. The proposed method, CEL, exploits both sides by giving each sample one embedding per class, then training with two additional losses. A class associative loss pulls candidate-set embeddings together and pushes non-candidate embeddings apart, while a prototype discriminative loss aligns the model's most confident candidate class with global class prototypes. Across six benchmarks, including four fine-grained image sets, CEL reports the best accuracy among twelve compared PLL and IDPLL methods.

What carries the argument

The key machinery is the class-wise embedding: instead of one feature vector per image, the model outputs a matrix $E_i \in \mathbb{R}^{q \times l}$, where row $j$ is the embedding of sample $x_i$ seen from class $j$, produced by an ML-Decoder encoder on top of a ResNet backbone. The class associative loss (CAL) uses cosine similarity to raise the average similarity of class-wise embeddings within the candidate set and lower it for pairs crossing the candidate boundary. The prototype discriminative loss (PDL) maintains a normalized prototype $Q_c$ for each class, updated only with embeddings of the highest-confidence candidate class, and then pulls that class's embedding toward $Q_c$ while pushing it away from all other prototypes. The two-stage schedule adds PDL only after $T_w$ epochs, on the ground that early prototypes are too unreliable to guide disambiguation.

What would settle it

Train CEL on one of the reported data sets (for example CUB200) with the prototype update in Eq. (6) replaced by a random candidate class, and compare final accuracy with the reported 68.60%. If the random-update control matches or beats the argmax-update version, the self-training assumption behind PDL is not doing the work the paper claims. A second check is a per-epoch ablation that adds PDL at different epochs; if adding it before epoch 250 does not hurt, the stated rationale for the two-stage schedule is falsified.

Watch

Extended reading notes

Core claim

The central claim is that the instance-dependence of noisy labels, usually treated only as a source of ambiguity, can be used as structured supervision. CEL equips each sample with class-wise embeddings, one vector per class, and imposes two geometric constraints: embeddings inside the candidate set should be mutually similar, embeddings between candidate and non-candidate classes should be dissimilar, and the embedding of the model's highest-confidence candidate class should sit close to that class's prototype while staying far from other prototypes. The paper reports that this combination gives first place on all six tested data sets, with the largest margins on fine-grained data such as CUB200 (68.60% versus 66.60% for the previous best method) and DOGS120 (78.18% versus 75.97%).

Load-bearing premise

The method assumes that the class the model is most confident about inside the candidate set is usually the true label, so accumulating that class's embedding into the class prototype builds a trustworthy global picture to guide later disambiguation.

Editorial extensions

If this is right

If CEL's claim holds, instance-dependent noisy labels should be treated as structured supervision rather than noise, and methods that ignore the relationship between noisy labels and image features are leaving accuracy on the table.
The ablation argues for a two-phase recipe: first learn representations that encode candidate-set structure with CAL, then disambiguate with global prototypes via PDL, with the paper attributing average gains of 1.22% and 0.71% to the two losses respectively.
The largest reported gaps on fine-grained data suggest that class-wise embeddings help most when candidate labels share a superclass and are visually similar, which is exactly the setting where ordinary PLL methods degrade.
The embedding-length experiments indicate that a length of 512 is a safe default across both small (32x32) and large (224x224) inputs, since oversizing dilutes features on small images and undersizing compresses features on large ones.
The win/tie/loss counts, 66 wins out of 72 paired comparisons at the 0.05 significance level, make the accuracy ranking a statistical claim and not just a point estimate.

Reading between the lines

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

The prototype update in Eq. (6) is a self-training loop: if the argmax within the candidate set is wrong early on, wrong embeddings enter the prototypes and PDL amplifies the error. The paper waits 250 epochs before adding PDL; a natural testable extension is a confidence-weighted or thresholded prototype update that softens this dependence.
The class associative loss encodes the prior that all candidate labels are semantically alike, which is true under instance-dependent noise but false under instance-independent partial labels. CEL may therefore transfer poorly to classic PLL benchmarks, and a simple experiment would be to run CEL on instance-independent candidate sets and compare with methods designed for that case.
Because CAL and PDL operate directly on the embedding geometry, the recipe is agnostic to the classifier head; a plausible extension is to test whether the same two losses give comparable gains with transformer backbones or higher-resolution inputs than the ResNet/ML-Decoder setup reported here.
The paper attributes the early training speed to CAL and the late-stage accuracy to PDL, but the ablation only measures final accuracy. A per-epoch ablation would directly test that attribution and, if it failed, would point to a different division of labor between the two losses.

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.

Desk Editor's Note

read the letter

Worth a look if you work on partial label learning. The core idea—giving each sample a separate embedding per class, then using those embeddings to enforce candidate-set similarity and prototype-based disambiguation—is new for IDPLL, and the results are consistently positive: CEL ranks first on all six benchmarks, with gains of 1–3 points over the previous best on fine-grained sets. The ablations show both losses contribute (CAL ~1.2%, PDL ~0.7% on average), and the significance tests are a nice addition. The comparison setup is reasonably even-handed: same backbones, same training schedule, baselines given their reported settings, and code is public. I believe the central claim.

Two real weaknesses. First, the prototype update in Eq. (6) is underspecified and internally inconsistent. The prose says you pick the highest-probability class within the candidate set, but the equation uses a global argmax with a membership test. If the implementation follows the equation, samples whose global argmax falls outside S_i never update any prototype, which biases coverage; if it follows the prose, the equation is miswritten. Either way, the reader can't tell what the actual loop does. Second, the self-training nature of the prototypes isn't probed. Prototypes are built from model predictions, and the paper never compares against ground-truth or random prototypes, so contamination from early wrong argmaxes is unquantified. Hyperparameters are also tuned on the same benchmarks, and all synthetic noise comes from one generation process (VALEN), so the generality of the gains is an open question.

These are fixable with additional experiments rather than fatal flaws. The empirical core is believable; the gains are consistent and the ablations support the contribution of each loss. The novelty claim ('first to create class-wise embeddings') is a bit overbroad since class-wise encoders exist in the cited multi-label literature, but the specific combination for IDPLL is new.

Who is this for? Anyone working on weakly supervised learning, specifically PLL/IDPLL. It deserves a serious referee: the method is clearly described (modulo the prototype issue), the experiments are thorough, and the code is public. Recommendation: send to peer review, and ask the authors to fix Eq. (6)/prose inconsistency and add a prototype-quality analysis before acceptance.

Referee Report

5 major / 4 minor

Summary. The paper proposes CEL, a method for instance-dependent partial label learning (IDPLL). For each sample, CEL produces class-wise embeddings, one per class, via an ML-Decoder head. It introduces two losses: a class associative loss (CAL) that pulls together embeddings of candidate labels and pushes apart embeddings of candidate and non-candidate labels, and a prototype discriminative loss (PDL) that aligns the embedding of the model's high-confidence candidate class with a class prototype and repels it from other prototypes. Prototypes are updated online from model predictions. Training is split into two stages, with PDL added after a warm-up of T_w=250 epochs. Experiments on six benchmarks against twelve methods report first place on all six data sets, with significance tests, ablations, parameter sensitivity plots, and a public code release.

Significance. If the method's gains are robust, CEL is a meaningful step for IDPLL: it is among the first to exploit per-class embeddings in this setting, it explicitly addresses the two-sided nature of instance-dependent noise, and the empirical evaluation is broad, with six benchmarks, twelve baselines, and public code. The reported improvements on fine-grained data sets (e.g., CUB200 from 66.60% to 68.60%, DOGS120 from 75.97% to 78.18%) are substantial. The main risks are technical: the prototype update rule appears inconsistent with the prose, the CAL objective as written contains a diagonal-term bias, and the self-training loop behind PDL is not diagnosed. The absence of a validation-based hyperparameter selection protocol also tempers the strength of the comparative claims.

major comments (5)

[Section 3.3, Eq. (3)] The sums over j,k ∈ S_i include the diagonal terms j=k, each of which contributes ⟨E_j,E_j⟩=1 to the numerator and 1 to the denominator. Since the class-wise embeddings are L2-normalized, this introduces a positive floor of 1/|S_i| into s_cal and prevents the first term of L_cal in Eq. (5) from being minimized to zero even under perfect candidate-set alignment. The pairwise similarity should be computed over ordered pairs with j≠k (or at least the diagonal should be excluded). Please correct Eq. (3) and confirm that the implementation matches the corrected definition.
[Section 3.4, Eq. (6)] The prose before Eq. (6) states that the authors select the class with the highest model output probability in the candidate label set, but Eq. (6) uses c = argmax(P_i) together with the condition c ∈ S_i. Under the displayed equation, samples whose global argmax falls outside S_i do not update any prototype, which both reduces prototype coverage and biases prototypes toward samples whose global argmax happens to be a candidate. Under the prose, the equation is miswritten. Please specify the exact selection rule used in the released code, correct the notation in Eq. (6) (and in Eqs. (7)-(8)), and clarify how the normalization in Eq. (6) is accumulated across updates.
[Section 3.4 and Table 4] Because prototypes are updated from the model's own argmax predictions, PDL creates a self-training loop: wrong early predictions contaminate prototypes, and Eq. (9) then reinforces those errors. The paper does not report any diagnostic of the quality of the high-confidence selections used in Eq. (6), nor does it compare prototypes built from predicted labels against ground-truth or random prototypes. Table 4 reports only the aggregate contribution of PDL (about 0.71% average), which cannot distinguish a genuine disambiguation gain from error amplification. Please add a plot or table showing selection precision/recall over training and at least one control experiment (e.g., ground-truth prototypes, fixed random prototypes) to quantify contamination.
[Section 4.1.3 and Table 2] The hyperparameters α, β, γ1, γ2 are selected by grid search on the same data sets whose test accuracies are reported in Table 2, and T_w=250 is fixed without a validation-based justification. Since the baselines use hyperparameters from their original papers, this asymmetric tuning can inflate the reported gains. Figure 5 shows sensitivity on two data sets, but the headline numbers are still best grid values on the test data. Please describe a validation split or other selection protocol, report the chosen hyperparameter values per data set, or provide full grid results for all benchmarks.
[Section 4.1.1] All experiments use a single synthetic IDPLL noise generation process, the one proposed by VALEN. Given that the paper motivates IDPLL by real-world annotation scenarios, the robustness of CAL and PDL to other instance-dependent noise models is unknown. Please either add experiments with at least one additional generation mechanism or discuss this limitation explicitly and temper the general claim of effectiveness.

minor comments (4)

[Table 2] The PICO row reports '67.80 ±-0.06%'; the standard deviation should be positive, so this is presumably a typo for 0.06%.
[Table 1] For CIFAR-100H, the reported average number of candidate labels is 3.41 while the rate is 0.6; for a 100-class problem, rate=0.6 would typically imply about 60 candidates, not 3.41. Please clarify what 'rate' means for CIFAR-100H and make the table internally consistent.
[Figure 5] The x-axis labels in panels (a)-(d) are visually misaligned (e.g., '1' and '2' appear at the bottom of panels (c) and (d) in a way that makes it hard to read the grid values). Please redraw the figure with clearer axis ticks.
[Abstract and Section 1] The claim of being 'the first to create class-wise embeddings for each sample' is too broad, because class-wise embeddings already appear in multi-label classification (e.g., Query2Label and ML-Decoder, both cited). The novelty claim should be scoped to partial label learning or IDPLL.

Circularity Check

2 steps flagged · score 6.0 of 10

Empirical claims on CIFAR-100 and CUB200 reduce to test-set hyperparameter/architecture selection; core loss derivation is otherwise self-contained.

fitted input called prediction [Section 4.3.2 (Parameters sensitivity), Fig. 5; Section 4.1.3 (Implementation details); Table 2]
"Fig. 5 shows the classification accuracy of our method CEL on benchmark data sets CIFAR-100 and CUB200 under different parameter settings. ... As illustrated in Fig. 5, when α is set to 0.5, β to 1, and γ1 to 1, the model achieves the best classification performance."

The same two benchmark test sets whose accuracies are reported in Table 2 are used to select α, β, γ1, and γ2; the paper describes no held-out validation split. The reported values 75.51% (CIFAR-100) and 68.60% (CUB200) are therefore the maxima of the swept parameter grid by construction, so the empirical 'prediction' on these datasets is a fitted selection rather than an independent test of the method.
fitted input called prediction [Section 4.3.3 (Length of the class-wise embedding), Fig. 6; Table 2]
"As shown in Fig. 6, for data set CIFAR-100, which has smaller image sizes (32×32), the classification accuracy of the model is higher when the length of the class-wise embedding is less than or equal to 512. ... Conversely, for the data set CUB200, which has larger image sizes (224×224), the classification accuracy is higher when the length of the class-wise embedding is greater than or equal to 512 ... Therefore, considering both cases, setting the class-wise embedding length to 512 is a good choice."

The class-wise embedding length l is selected by inspecting classification accuracy on exactly the same CIFAR-100 and CUB200 test sets that Table 2 later reports as results. No validation split is mentioned. The reported test accuracies for these two datasets are thus partly produced by the selection criterion, not independently predicted.

full rationale

The derivation of the two proposed losses is self-contained: CAL (Eqs. 3–5) is defined directly from candidate/non-candidate set relationships, PDL (Eqs. 6–9) is defined from model predictions and moving-average prototypes, and the final accuracy is measured on held-out test labels. The prototype update in Eq. (6) creates a self-referential training loop, but that is a training-dynamics concern, not a logical reduction of the accuracy claim to an input. No load-bearing self-citation, imported uniqueness theorem, or ansatz-smuggling via citation is present. The concrete circularity is in the evaluation protocol: hyperparameters α, β, γ1, γ2 and the embedding length l are chosen by maximizing accuracy on the same CIFAR-100 and CUB200 benchmark test sets that are then reported as the method's performance, with no separate validation set described. Those particular reported numbers are therefore maxima over the searched configurations by construction. The other four benchmark results and the general algorithmic contribution retain independent content, so the paper is only partially circular rather than fully reducible.

Assumptions & free parameters 6 free parameters · 5 assumptions · 0 invented entities

The central claim rests on an empirical training procedure with six tuned hyperparameters and four domain assumptions about IDPLL structure and prototype reliability. No new physical or mathematical entities are introduced; class prototypes are an algorithmic construct from prior methods. The main cost is that the method's success is tied to how well the assumed similarity structure and prototype confidence hold for a given dataset.

free parameters (6)

alpha (CAL weight) = 0.5
Trade-off weight for class associative loss in Eqs. (10-11); selected from {0.1, 0.5, 1} via sensitivity analysis on CIFAR-100 and CUB200.
beta (PDL weight) = 1
Trade-off weight for prototype discriminative loss in Eq. (11); selected from {0.1, 0.5, 1} via sensitivity analysis.
gamma1 (CAL balance) = 1
Balances pull-close and push-away terms in Eq. (5); selected from {0.5, 1, 2, 5}.
gamma2 (PDL balance) = 1 or 2, dataset dependent
Balances prototype alignment and separation in Eq. (9); best value differs between CUB200 (1) and CIFAR-100 (2), so a single value is not fixed by the paper.
T_w (first-stage epochs) = 250
Number of epochs before PDL is added; set to half of 500 total epochs without a dedicated study.
l (class-wise embedding length) = 512
Length of each class-wise embedding; chosen from {128, 256, 512, 768, 1024} based on a sensitivity experiment on CIFAR-100 and CUB200.

assumptions (5)

domain assumption In IDPLL, labels inside the candidate set are semantically similar to each other and labels outside it are dissimilar.
This is the premise of the class associative loss in Section 3.3, Eqs. (3)-(5). If real IDPLL noise does not follow this similarity structure, CAL would enforce misleading geometry.
domain assumption The argmax model prediction within the candidate set is reliable enough to build class prototypes.
Section 3.4, Eq. (6) accumulates only the highest-probability candidate embedding into prototype Q_c; early wrong predictions would poison the prototypes.
domain assumption Class-wise embeddings produced by ML-Decoder capture per-class semantic information needed for the two losses.
Section 3.1 and implementation rely on ML-Decoder to produce meaningful E_i in R^{q x l}; if these embeddings are not class-discriminative, CAL and PDL have no reliable signal.
domain assumption The VALEN generation process produces instance-dependent candidate sets representative of real-world IDPLL.
All six datasets use VALEN noise (Section 4.1.1); the claim of practical relevance depends on this protocol transferring to real ambiguous labels.
domain assumption Gradient descent on the composite loss generalizes to held-out data.
The paper provides no generalization bound or theoretical analysis; effectiveness is shown only empirically.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Mixed Blessing: Class-Wise Embedding guided Instance-Dependent Partial Label Learning." pith.science (2026). https://pith.science/paper/TCNI73FO

@misc{pith2026241205029,
  author       = {Pith},
  title        = {Pith review of: Mixed Blessing: Class-Wise Embedding guided Instance-Dependent Partial Label Learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/TCNI73FO}},
  note         = {Machine review of arXiv:2412.05029}
}

read the original abstract

In partial label learning (PLL), every sample is associated with a candidate label set comprising the ground-truth label and several noisy labels. The conventional PLL assumes the noisy labels are randomly generated (instance-independent), while in practical scenarios, the noisy labels are always instance-dependent and are highly related to the sample features, leading to the instance-dependent partial label learning (IDPLL) problem. Instance-dependent noisy label is a double-edged sword. On one side, it may promote model training as the noisy labels can depict the sample to some extent. On the other side, it brings high label ambiguity as the noisy labels are quite undistinguishable from the ground-truth label. To leverage the nuances of IDPLL effectively, for the first time we create class-wise embeddings for each sample, which allow us to explore the relationship of instance-dependent noisy labels, i.e., the class-wise embeddings in the candidate label set should have high similarity, while the class-wise embeddings between the candidate label set and the non-candidate label set should have high dissimilarity. Moreover, to reduce the high label ambiguity, we introduce the concept of class prototypes containing global feature information to disambiguate the candidate label set. Extensive experimental comparisons with twelve methods on six benchmark data sets, including four fine-grained data sets, demonstrate the effectiveness of the proposed method. The code implementation is publicly available at https://github.com/Yangfc-ML/CEL.

Figures

Figures reproduced from arXiv: 2412.05029 by the authors.

**Figure 2.** The classification accuracy curves of PLL method [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗

**Figure 3.** Illustration of our method CEL. Our model consists of three modules: the backbone [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗

Figures from the paper (3 more)

**Figure 4.** Figure 4: Classification accuracy curves of all methods on [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]

**Figure 5.** Figure 5: Parameters sensitivity of our method CEL. (a) - (d) represent the classification accuracy of our method on benchmark [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]

**Figure 6.** Figure 6: Classification accuracy of different lengths of class [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

43 extracted references · 41 canonical work pages

[1]

Timothée Cour, Benjamin Sapp, and Ben Taskar. 2011. Learning from Partial Labels. Journal of Machine Learning Research 12 (2011), 1501–1536

work page 2011
[2]

Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. 2009. Ima- geNet: A large-scale hierarchical image database. In 2009 IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR 2009) . 248–255

work page 2009
[3]

Lei Feng and Bo An. 2018. Leveraging Latent Label Distributions for Partial Label Learning. In International Joint Conference on Artificial Intelligence . 2107–2113

work page 2018
[4]

Lei Feng and Bo An. 2019. Partial Label Learning by Semantic Difference Maxi- mization. In International Joint Conference on Artificial Intelligence . 2294–2300

work page 2019
[5]

Lei Feng, Jiaqi Lv, Bo Han, Miao Xu, Gang Niu, Xin Geng, Bo An, and Masashi Sugiyama. 2020. Provably Consistent Partial-Label Learning. In Advances in Neural Information Processing Systems 33

work page 2020
[6]

Girshick

Kaiming He, Haoqi Fan, Yuxin Wu, Saining Xie, and Ross B. Girshick. 2020. Momentum Contrast for Unsupervised Visual Representation Learning. In 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2020 . 9726–9735

work page 2020
[7]

Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. 2016. Deep Residual Learning for Image Recognition. In 2016 IEEE Conference on Computer Vision and Pattern Recognition, CVPR 2016 . 770–778

work page 2016
[8]

Shuo He, Guowu Yang, and Lei Feng. 2023. Candidate-aware Selective Disam- biguation Based On Normalized Entropy for Instance-dependent Partial-label Learning. In IEEE/CVF International Conference on Computer Vision . 1792–1801

work page 2023

Show all 43 references

[9]

Eyke Hüllermeier and Jürgen Beringer. 2006. Learning from ambiguously labeled examples. Intelligent Data Analysis 10, 5 (2006), 419–439. http://content.iospress. com/articles/intelligent-data-analysis/ida00259

2006
[10]

Yuheng Jia, Xiaorui Peng, Ran Wang, and Min-Ling Zhang. 2024. Long-Tailed Partial Label Learning by Head Classifier and Tail Classifier Cooperation. In Thirty-Eighth AAAI Conference on Artificial Intelligence . 12857–12865

2024
[11]

Yuheng Jia, Fuchao Yang, and Yongqiang Dong. 2023. Partial Label Learning with Dissimilarity Propagation guided Candidate Label Shrinkage. In Advances in Neural Information Processing Systems 36

2023
[12]

Jiahao Jiang, Yuheng Jia, Hui Liu, and Junhui Hou. 2024. FairMatch: Promoting Partial Label Learning by Unlabeled Samples. In ACM SIGKDD Conference on Knowledge Discovery and Data Mining . 1269–1278

2024
[13]

Aditya Khosla, Nityananda Jayadevaprakash, Bangpeng Yao, and Li Fei-Fei. 2011. Novel Dataset for Fine-Grained Image Categorization. In First Workshop on Fine- Grained Visual Categorization, IEEE Conference on Computer Vision and Pattern Recognition

2011
[14]

Jonathan Krause, Michael Stark, Jia Deng, and Li Fei-Fei. 2013. 3D Object Repre- sentations for Fine-Grained Categorization. In 2013 IEEE International Conference on Computer Vision Workshops. 554–561

2013
[15]

Alex Krizhevsky. 2009. Learning Multiple Layers of Features from Tiny Images. (2009)

2009
[16]

Jack Lanchantin, Tianlu Wang, Vicente Ordonez, and Yanjun Qi. 2021. Gen- eral Multi-Label Image Classification With Transformers. In IEEE Conference on Computer Vision and Pattern Recognition, CVPR 2021 . 16478–16488

2021
[17]

Changchun Li, Ximing Li, and Jihong Ouyang. 2020. Learning with Noisy Partial Labels by Simultaneously Leveraging Global and Local Consistencies. In ACM International Conference on Information and Knowledge Management . 725–734

2020
[18]

Dietterich

Li-Ping Liu and Thomas G. Dietterich. 2012. A Conditional Multinomial Mixture Model for Superset Label Learning. In Advances in Neural Information Processing Systems 25: 26th Annual Conference on Neural Information Processing Systems . 557–565

2012
[19]

Shilong Liu, Lei Zhang, Xiao Yang, Hang Su, and Jun Zhu. 2021. Query2Label: A Simple Transformer Way to Multi-Label Classification. CoRR abs/2107.10834 (2021). arXiv:2107.10834

2021 arXiv
[20]

Jie Luo and Francesco Orabona. 2010. Learning from Candidate Labeling Sets. In Advances in Neural Information Processing Systems 23: 24th Annual Conference on Neural Information Processing Systems . 1504–1512

2010
[21]

Jiaqi Lv, Biao Liu, Lei Feng, Ning Xu, Miao Xu, Bo An, Gang Niu, Xin Geng, and Masashi Sugiyama. 2024. On the Robustness of Average Losses for Partial-Label Learning. IEEE Trans. Pattern Anal. Mach. Intell. 46, 5 (2024), 2569–2583

2024
[22]

Jiaqi Lv, Miao Xu, Lei Feng, Gang Niu, Xin Geng, and Masashi Sugiyama. 2020. Progressive Identification of True Labels for Partial-Label Learning. In Interna- tional Conference on Machine Learning, ICML 2020 . 6500–6510

2020
[23]

Blaschko, and Andrea Vedaldi

Subhransu Maji, Esa Rahtu, Juho Kannala, Matthew B. Blaschko, and Andrea Vedaldi. 2013. Fine-Grained Visual Classification of Aircraft. CoRR abs/1306.5151 (2013). arXiv:1306.5151

2013 arXiv
[24]

Congyu Qiao, Ning Xu, and Xin Geng. 2023. Decompositional Generation Process for Instance-Dependent Partial Label Learning. In The Eleventh International Conference on Learning Representations, ICLR 2023

2023
[25]

Tal Ridnik, Gilad Sharir, Avi Ben-Cohen, Emanuel Ben Baruch, and Asaf Noy
[26]

Shiyu Tian, Hongxin Wei, Yiqun Wang, and Lei Feng. 2024. CroSel: Cross Selec- tion of Confident Pseudo Labels for Partial-Label Learning. In IEEE/CVF Confer- ence on Computer Vision and Pattern Recognition, CVPR 2024 . 19479–19488

2024
[27]

Yingjie Tian, Xiaotong Yu, and Saiji Fu. 2023. Partial label learning: Taxonomy, analysis and outlook. Neural Networks 161 (2023), 708–734

2023
[28]

Catherine Wah, Steve Branson, Peter Welinder, Pietro Perona, and Serge Be- longie. 2011. The Caltech-UCSD Birds-200-2011 Dataset. California Institute of Technology (2011)

2011
[29]

Deng-Bao Wang, Min-Ling Zhang, and Li Li. 2022. Adaptive Graph Guided Disambiguation for Partial Label Learning. IEEE Trans. Pattern Anal. Mach. Intell. 44, 12 (2022), 8796–8811

2022
[30]

Haobo Wang, Ruixuan Xiao, Yixuan Li, Lei Feng, Gang Niu, Gang Chen, and Junbo Zhao. 2022. PiCO: Contrastive Label Disambiguation for Partial Label Learning. In The Tenth International Conference on Learning Representations, ICLR 2022

2022
[31]

Belongie

Xiu-Shen Wei, Yi-Zhe Song, Oisin Mac Aodha, Jianxin Wu, Yuxin Peng, Jinhui Tang, Jian Yang, and Serge J. Belongie. 2022. Fine-Grained Image Analysis With Deep Learning: A Survey. IEEE Trans. Pattern Anal. Mach. Intell. 44, 12 (2022), 8927–8948

2022
[32]

Hongwei Wen, Jingyi Cui, Hanyuan Hang, Jiabin Liu, Yisen Wang, and Zhouchen Lin. 2021. Leveraged Weighted Loss for Partial Label Learning. In Proceedings of the 38th International Conference on Machine Learning, ICML 2021 , Vol. 139. 11091–11100

2021
[33]

Dong-Dong Wu, Deng-Bao Wang, and Min-Ling Zhang. 2022. Revisiting Consis- tency Regularization for Deep Partial Label Learning. In International Conference on Machine Learning, ICML 2022 , Vol. 162. 24212–24225

2022
[34]

Dong-Dong Wu, Deng-Bao Wang, and Min-Ling Zhang. 2024. Distilling Reliable Knowledge for Instance-Dependent Partial Label Learning. InThirty-Eighth AAAI Conference on Artificial Intelligence. 15888–15896

2024
[35]

Shiyu Xia, Jiaqi Lv, Ning Xu, and Xin Geng. 2022. Ambiguity-Induced Con- trastive Learning for Instance-Dependent Partial Label Learning. In Thirty-First International Joint Conference on Artificial Intelligence . 3615–3621

2022
[36]

Shiyu Xia, Jiaqi Lv, Ning Xu, Gang Niu, and Xin Geng. 2023. Towards Effective Visual Representations for Partial-Label Learning. In IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2023 . 15589–15598

2023
[37]

Ning Xu, Biao Liu, Jiaqi Lv, Congyu Qiao, and Xin Geng. 2023. Progressive Purifi- cation for Instance-Dependent Partial Label Learning. In International Conference on Machine Learning, ICML 2023 . 38551–38565

2023
[38]

Ning Xu, Congyu Qiao, Xin Geng, and Min-Ling Zhang. 2021. Instance- Dependent Partial Label Learning. In Advances in Neural Information Processing Systems 34. 27119–27130

2021
[39]

Zinan Zeng, Shijie Xiao, Kui Jia, Tsung-Han Chan, Shenghua Gao, Dong Xu, and Yi Ma. 2013. Learning by Associating Ambiguously Labeled Images. In 2013 IEEE Conference on Computer Vision and Pattern Recognition . 708–715

2013
[40]

Min-Ling Zhang and Fei Yu. 2015. Solving the Partial Label Learning Prob- lem: An Instance-Based Approach. In International Joint Conference on Artificial Intelligence. 4048–4054

2015
[41]

Min-Ling Zhang, Bin-Bin Zhou, and Xu-Ying Liu. 2016. Partial Label Learning via Feature-Aware Disambiguation. In ACM SIGKDD Conference on Knowledge Discovery and Data Mining . 1335–1344

2016
[42]

Bo Zhao, Jiashi Feng, Xiao Wu, and Shuicheng Yan. 2017. A survey on deep learning-based fine-grained object classification and semantic segmentation. Int. J. Autom. Comput. 14, 2 (2017), 119–135

2017
[2023]

InIEEE/CVF Winter Conference on Applications of Computer Vision, W ACV 2023

ML-Decoder: Scalable and Versatile Classification Head. InIEEE/CVF Winter Conference on Applications of Computer Vision, W ACV 2023. 32–41

2023

Pith tools

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

[1] [1]

Timothée Cour, Benjamin Sapp, and Ben Taskar. 2011. Learning from Partial Labels. Journal of Machine Learning Research 12 (2011), 1501–1536

work page 2011

[2] [2]

Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. 2009. Ima- geNet: A large-scale hierarchical image database. In 2009 IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR 2009) . 248–255

work page 2009

[3] [3]

Lei Feng and Bo An. 2018. Leveraging Latent Label Distributions for Partial Label Learning. In International Joint Conference on Artificial Intelligence . 2107–2113

work page 2018

[4] [4]

Lei Feng and Bo An. 2019. Partial Label Learning by Semantic Difference Maxi- mization. In International Joint Conference on Artificial Intelligence . 2294–2300

work page 2019

[5] [5]

Lei Feng, Jiaqi Lv, Bo Han, Miao Xu, Gang Niu, Xin Geng, Bo An, and Masashi Sugiyama. 2020. Provably Consistent Partial-Label Learning. In Advances in Neural Information Processing Systems 33

work page 2020

[6] [6]

Girshick

Kaiming He, Haoqi Fan, Yuxin Wu, Saining Xie, and Ross B. Girshick. 2020. Momentum Contrast for Unsupervised Visual Representation Learning. In 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2020 . 9726–9735

work page 2020

[7] [7]

Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. 2016. Deep Residual Learning for Image Recognition. In 2016 IEEE Conference on Computer Vision and Pattern Recognition, CVPR 2016 . 770–778

work page 2016

[8] [8]

Shuo He, Guowu Yang, and Lei Feng. 2023. Candidate-aware Selective Disam- biguation Based On Normalized Entropy for Instance-dependent Partial-label Learning. In IEEE/CVF International Conference on Computer Vision . 1792–1801

work page 2023