LoD: Loss-difference OOD Detection by Intentionally Label-Noisifying Unlabeled Wild Data

Chuanxing Geng; Dong Liang; Pong C. Yuen; Qifei Li; Songcan Chen; Xinrui Wang

REVIEW 4 major objections 6 minor 45 references

LoD: Loss-difference OOD Detection by Intentionally Label-Noisifying Unlabeled Wild Data

T0 review · 4 major / 6 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read The paper claims that intentionally labeling all unlabeled wild data as a single fake extra class makes OOD samples separable from ID samples by K-means on training losses, with no threshold required.

desk verdict The empirical recipe is a genuine new angle and the results are strong; the theory is not load-bearing as written, so treat the paper as an empirical contribution until the proof is repaired. read the letter →

arxiv 2505.12952 v1 pith:SWOPBT3N submitted 2025-05-19 cs.LG stat.ML

Chuanxing Geng , Qifei Li , Xinrui Wang , Dong Liang , Songcan Chen , Pong C. Yuen This is my paper · ORCID

classification cs.LGstat.ML

keywords out-of-distributiondetectionunlabeledwilddatalabel-noiselearningearlyK-meansfilteringthreshold-freeHubercontaminationmodellossgap

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 tries to show that a known difficulty in out-of-distribution (OOD) detection — how to use unlabeled wild data that mixes in-distribution (ID) and OOD images — can be handled by deliberately labeling all wild data as a single extra class, the K+1-th class. The labeling converts ID samples in the wild into label-noise samples and OOD samples into label-clean samples, so the network's own training losses separate the two: noisy ID samples keep higher, more fluctuating losses while clean OOD samples settle to lower losses. That separation is strong enough that plain K-means clustering on mean training losses picks out the OOD samples, eliminating the need for a threshold. The paper backs this with a theoretical loss-gap bound and reports false-positive rates near 0% on standard benchmarks while keeping competitive ID accuracy. A sympathetic reader would care because threshold-free filtering and joint ID/OOD dominance remove two design obstacles in wild-data OOD detection.

What carries the argument

The load-bearing mechanism is intentional label-noisification: assigning the single K+1-th class to all unlabeled wild data, which converts the OOD-filtering problem into a label-noise learning problem with a controllable noise ratio. The work it does is to produce the early-learning loss asymmetry — clean OOD samples fit quickly, noisy ID samples do not — and the resulting mean-loss gap that two-cluster K-means exploits. The theoretical engine is Proposition 1, which transfers an early-learning guarantee from a two-class Gaussian model to a lower bound on the difference between mean cross-entropy losses of ID and OOD wild samples; the clustering step replaces threshold selection with the distance rule $d_1 < d_2$. A useful named model is the Huber contamination model $P_{\mathrm{wild}} = (1-\pi) P_{\mathrm{in}} + \pi P_{\mathrm{out}}$, which formalizes the wild mixture.

What would settle it

Train LoD on a wild set whose OOD classes are chosen to be nearest neighbours of the ID classes, then plot the per-sample mean-loss histograms for true ID and true OOD samples. If the two histograms overlap so strongly that two-cluster K-means purity on the wild set falls to chance level, or if the measured mean-loss gap is not positive, the central claim is falsified. A simpler observational check: on the paper's own TinyImageNet hard benchmark, the reported FPR95 around 45% already indicates imperfect cluster separation, so measuring whether cluster accuracy tracks the Proposition 1 bound would settle the issue.

Watch

Extended reading notes

Core claim

The central claim is that OOD filtering in unlabeled wild data can be reformulated as a controllable label-noise learning problem. Label every sample in the wild set as class K+1 and train a K+1-way classifier together with clean K-class labeled ID data. During the early-learning phase, the K+1-labeled ID samples behave like mislabeled data: their cross-entropy losses stay higher and fluctuate more, whereas K+1-labeled OOD samples behave like clean data and converge to low losses. Proposition 1 formalizes this as a gap in mean losses bounded below by $1 - 2 e^{-\theta^T v + \frac{1}{2}\|\theta\|^2 \delta^2} - O(R/\sqrt{n})$, so with enough samples the two populations separate; K-means on per-sample mean losses with two clusters then identifies OOD without any threshold. The framework then trains a binary OOD detector on the filtered OOD candidates plus labeled ID data. The empirical claim is that this yields near-zero FPR95 on standard benchmarks and large margins over prior wild-data methods on harder same-dataset benchmarks.

Load-bearing premise

The method stands or falls on the assumption that real out-of-distribution images, once labeled as the fake extra class, are consistently easier for the deep network to fit than real in-distribution images that were deliberately mislabeled as that same class; the simplified two-Gaussian proof does not by itself guarantee that this ordering holds for a deep K+1 classifier on arbitrary wild mixtures.

Editorial extensions

If this is right

If the loss-gap premise holds, OOD filtering from wild data needs no threshold: a fixed two-cluster K-means on mean training losses is sufficient.
Because wild OOD samples are correctly labeled as the extra class and jointly dominate learning with labeled ID data, the model-bias problem that makes networks fit OOD as ID is reduced.
The label-noise ratio is controllable through the batch ratio $|B_{\mathrm{train}}^{\mathrm{in}}| : |B_{\mathrm{wild}}|$, and the paper reports that larger ratios improve filtering performance.
The framework is data-centric and architecture-agnostic, working across WideResNet-40-2, ResNet18, and ResNet34 without modifying network structure.
On hard benchmarks where ID and OOD come from the same dataset, the paper reports substantially lower FPR95 than prior wild-data methods, with the largest gains on CIFAR10-based splits.

Reading between the lines

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

Editorial inference: the same loss-gap mechanism should transfer to other training objectives that exhibit early-learning behavior, such as symmetric losses or robust cross-entropy variants, provided the asymmetry (ID noisy, OOD clean) is preserved; testing this would clarify whether the K+1 relabeling itself is the essential ingredient.
Editorial inference: the threshold-free property depends on the mean-loss distributions being bimodal and separable; on the hardest reported benchmark, TinyImageNet, FPR95 remains around 45%, so the universal no-threshold promise is bounded by how distinguishable OOD is from ID in feature space.
Editorial inference: because the noise ratio is controlled by batch composition, an online stopping rule based on cluster separation (e.g., a bimodality index or silhouette score) could make the early-learning window self-terminating instead of fixed at a chosen epoch count.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note

read the letter

Colleague,

Read LoD. The one thing to know: the core move is better than I expected from the title. Label every wild sample as class K+1, train a K+1 classifier, then K-means the per-sample mean losses to separate ID from OOD without thresholds. That genuinely reframes unlabeled-wild OOD filtering as label-noise learning, and the paper shows it works across standard and hard benchmarks with consistent gains over SAL, WOODS, OE, and Energy. This is the paper's real value and it should be preserved.

What is new: the deliberate K+1 relabeling is not in the cited baselines, and using loss-curve clustering rather than a threshold is a clean way around the threshold dilemma. The authors also curated hard benchmarks (same dataset, different classes) and included experiments across networks, ratios, and unseen OOD sets. That is a solid empirical package. Code is linked. I would believe the method works in the tested settings, though I would want error bars before trusting numbers like 0.00 FPR95: several are exact zeros, and no standard deviations or per-split intervals are reported.

Where it softens: the theoretical section does not establish what the paper claims. Eq. (5) decomposes the loss into three terms; Eq. (6) writes the gradient update as θ_{t+1} = θ_t − η(∇L_N1 + ∇L_N2), dropping the OOD term ∇L_N3. That is a concrete error, and the "joint dominance" conclusion does not follow from the displayed equation. Proposition 1 is worse. It is derived from the two-class Gaussian model with symmetric label flipping (Eq. 7), where clean and noisy samples have opposite labels and opposite logit signs. LoD's actual objective, Eq. (2), assigns both ID and OOD wild samples the same K+1 label; neither group receives an opposite label. So the inequalities l_clean ≤ e^{−θ^T x} and l_noise ≥ 1 − e^{−θ^T x} do not apply to the K+1 cross-entropy loss. The stress-test note is right: the proof proves a statement about a different labeling scheme, and even inside that scheme the lower bound's positivity is not shown. This is not a missing step; the object being analyzed is not LoD's loss.

Minor issues: the ratio ablation supports the gradient intuition but does not rescue it, and the "long early-learning" explanation for 100–500 epochs is plausible but hand-wavy. The citations to the label-noise literature are appropriate.

Bottom line: this paper deserves serious refereeing because the empirical method is novel, simple, and probably useful to the OOD detection and label-noise communities. But the authors should be asked to either repair the proof or remove the formal claim and present Proposition 1 as intuition. I would accept a revised version with error bars and a corrected theory section.

Send it out.

Referee Report

4 major / 6 minor

Summary. The manuscript proposes LoD, a two-stage OOD detection framework that uses unlabeled wild data. In the filtering stage, all wild samples are intentionally labeled as a single K+1-th class and trained together with labeled ID data in a fully supervised K+1-class setting, so that OOD wild samples become effectively clean labels while ID wild samples become conflicting labels. The mean cross-entropy losses of wild samples over training are then clustered with K-means into two groups, ID (high loss) and OOD (low loss), removing the need for a manually tuned threshold. In the detection stage, the filtered OOD samples and labeled ID data train a binary OOD detector. The paper argues that this reformulation addresses model bias and threshold selection, provides a theoretical analysis of a claimed loss gap, and reports extensive experiments on standard and hard benchmarks.

Significance. If the loss-ordering claim holds, LoD is a simple, data-centric, threshold-free approach to wild-data OOD detection, and the paper has several strengths: the problem reformulation is clear and elegant, the experimental evaluation is extensive (standard and hard benchmarks, unseen OOD datasets, multiple network architectures, and ratio/epoch ablations), and the reported gains over OE, Energy, WOODS, and SAL are consistent. I also credit the paper for curating harder same-source benchmarks. However, the theoretical foundation as written does not establish the central loss-gap guarantee: the proof models a different binary label-noise setting and the displayed gradient update omits one of the three terms. The empirical contribution is strong, but the advertised theory needs substantial repair.

major comments (4)

[Sec. 4.1, Eq. (5)–(6)] The gradient update in Eq. (6) drops the OOD term. Eq. (5) decomposes the loss into three parts with gradients ∇L_N1, ∇L_N2, and ∇L_N3, but Eq. (6) writes θ_{t+1} = θ_t − η(∇L_N1 + ∇L_N2), omitting ∇L_N3. Consequently the displayed update contradicts the decomposition and cannot support the claim that labeled ID data and OOD wild data jointly dominate learning. Even if the term is reinserted, the sample-size argument |B_train_in| > |B_wild| ≥ N2 does not by itself imply that ∇L_N3 is non-negligible, since gradient magnitudes depend on the model state, not only on batch counts.
[Sec. 4.2 and Appendix A, Proposition 1] The proposition is proved in a two-class Gaussian model with symmetric label noise, where a clean sample has log-likelihood log(1+e^{−θ^T x}) and a noisy sample has log(1+e^{+θ^T x}). In LoD's actual objective, Eq. (2), both ID and OOD wild samples are assigned the same observed label y_{K+1}; there is no binary sign flip and the ID wild samples are not labeled with an opposite binary class. The inequalities l_clean ≤ e^{−θ^T(v+z)} and l_noise ≥ 1−e^{−θ^T(v+z)} therefore do not apply to the K+1-class cross-entropy loss. The proof establishes a statement about a different training objective, not about the LoD training loss.
[Sec. 4.2, Proposition 1 and Eq. (9)] Even within the binary model, the claimed lower bound is not shown to be positive. The right-hand side 1 − 2e^{−θ^T v + (1/2)||θ||^2 δ^2} − O(R/√n) can be negative unless θ^T v is sufficiently large relative to ||θ||^2 δ^2; Lemma 1 only guarantees the normalized gradient direction, not a lower bound on θ^T v itself. In addition, the proposition and its proof use δ^2 while Eq. (9) derives the Gaussian expectation with σ^2, so the notation conflates the label-noise parameter with the Gaussian variance. Without a positivity argument, the proposition cannot justify the K-means separation used in Eq. (3).
[Tables 1 and 2] The main empirical tables report exact 0.00 FPR95 and 100.00 AUROC values for many settings, but no error bars, standard deviations, or per-seed results are provided in the main text; Appendix B mentions averaging over five splits for hard benchmarks, yet the main tables do not report the spread. Given the saturation at 0/100, error bars are necessary to assess whether the differences among methods are meaningful and whether the 0.00 values are stable across runs.

minor comments (6)

[Appendix A, Eq. (9)] The derivation uses E[e^{−θ^T z}] = e^{(1/2)||θ||^2 σ^2}, but Proposition 1 and the text use δ^2; please make the notation consistent.
[Figure 4 caption] The caption contains a typo: 'different rations' should be 'different ratios'.
[Algorithm 1, line 6] The loss-recording step V ← V ∪ {l_i | i ∈ (1, |B_wild|)} is underspecified: it should state how losses are associated with individual samples in D_wild when batches are sampled with replacement, and whether each sample's loss is recorded once per epoch.
[Appendix C and D captions] The captions of Table 4 and Table 5 refer to 'Table 2 and Table 3' in a way that collides with the main-text numbering; please renumber the appendix tables and references.
[Sec. 4.2, Lemma 1/Proposition 1] The phrase 'with high probability' is used without a quantitative probability in either Lemma 1 or Proposition 1; please state the failure probability or clarify the o(1) convention.
[Sec. 3.2, Remark] Calling OOD wild samples 'label-clean' under the synthetic K+1-th class is conceptually nonstandard, since every wild sample receives the same artificially assigned label; a brief clarification of what 'clean' means in this synthetic-label setting would help readers.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: LoD's loss-gap mechanism is imported from external label-noise learning theory and evaluated on held-out OOD data.

full rationale

The paper's central derivation chain is not self-referential. The method intentionally labels all unlabeled wild data as a K+1-th class, converting ID wild samples into label-noise samples and OOD wild samples into label-clean samples. The claimed loss ordering between these two groups is then justified by the early-learning phenomenon imported from Liu et al. (2020a) and Yue and Jha (2024), both external works with no author overlap with the present paper. The K-means filtering step is unsupervised on the wild data: it clusters per-sample mean training losses into two groups and assigns the lower-loss cluster to OOD, which is a method design rather than a fitted parameter renamed as a prediction. The OOD detector is trained on the resulting candidate OOD set and evaluated on held-out OOD test sets, so the main empirical results are not equivalent to the training objective. The theoretical Proposition 1 has a known modeling gap: it is proved in a two-class symmetric-label-noise Gaussian setting and may not directly apply to the K+1-class softmax objective used in LoD, and Eq. (6) omits the OOD gradient term. However, these are correctness or rigor concerns, not circularity. There is no self-citation load-bearing step, no fitted input called a prediction, and no derivation that reduces to its own conclusion by definition. Therefore the paper receives a circularity score of 0.

Assumptions & free parameters 3 free parameters · 4 assumptions · 1 invented entities

The method's central claims rest on a small set of imported empirical phenomena and hand-chosen hyperparameters. No new physical entities are introduced; the only invented construct is the K+1 pseudo-class. The two-Gaussian analysis is a simplified stand-in for the deep-network setting, and the positivity of the claimed loss gap is not established.

free parameters (3)

Batch ratio |B_train_in|:|B_wild| = 3:1
Selected after sweeping ratios 1:6, 1:3, 1:1, 3:1, 6:1 on benchmarks (Figure 4, Tables 7-8); not derived from theory.
Training epochs T = 100
The epoch study shows consistent gains from 100 to 500 (Figure 5, Table 9), but 100 is used for all experiments for efficiency; this is a hand-chosen hyperparameter.
Initial learning rates = 0.01 and 0.001
Set in Section 5.1 following prior works; hand-chosen, not derived from the theoretical analysis.

assumptions (4)

domain assumption Huber contamination model P_wild = (1-π)P_in + πP_out
Used in Eq. (1) to model wild data mixture; standard in the OOD literature but unverified for real deployments.
domain assumption Early-learning succeeds in label-noise training (Lemma 1 from Liu et al. 2020a)
The entire loss-gap argument imports this empirical/theoretical phenomenon; it is not re-derived for the K+1 setting.
ad hoc to paper Two-Gaussian model with label-noise ratio Δ is representative of the K+1 deep-network wild-data setting
Proposition 1 is proved only in this simplified setting (Section 4.2), but stated generally.
ad hoc to paper OOD samples under the K+1 label are learned faster than ID samples mislabeled as K+1
Used in the K-means rule Eq. (3) to assign the higher-mean-loss cluster to ID; no proof is given that this ordering holds for arbitrary OOD.

invented entities (1)

K+1-th pseudo-class 'wild' label
purpose: Container label for all unlabeled wild data; makes OOD samples label-clean and ID samples label-noise during filtering.
This is a construct introduced by the paper. Its value is only evidenced by downstream benchmark results; there is no external falsifiable handle.

how reviews work

0 comments

Cite this review

Pith. "Pith review of LoD: Loss-difference OOD Detection by Intentionally Label-Noisifying Unlabeled Wild Data." pith.science (2026). https://pith.science/paper/SWOPBT3N

@misc{pith2026250512952,
  author       = {Pith},
  title        = {Pith review of: LoD: Loss-difference OOD Detection by Intentionally Label-Noisifying Unlabeled Wild Data},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/SWOPBT3N}},
  note         = {Machine review of arXiv:2505.12952}
}

read the original abstract

Using unlabeled wild data containing both in-distribution (ID) and out-of-distribution (OOD) data to improve the safety and reliability of models has recently received increasing attention. Existing methods either design customized losses for labeled ID and unlabeled wild data then perform joint optimization, or first filter out OOD data from the latter then learn an OOD detector. While achieving varying degrees of success, two potential issues remain: (i) Labeled ID data typically dominates the learning of models, inevitably making models tend to fit OOD data as IDs; (ii) The selection of thresholds for identifying OOD data in unlabeled wild data usually faces dilemma due to the unavailability of pure OOD samples. To address these issues, we propose a novel loss-difference OOD detection framework (LoD) by \textit{intentionally label-noisifying} unlabeled wild data. Such operations not only enable labeled ID data and OOD data in unlabeled wild data to jointly dominate the models' learning but also ensure the distinguishability of the losses between ID and OOD samples in unlabeled wild data, allowing the classic clustering technique (e.g., K-means) to filter these OOD samples without requiring thresholds any longer. We also provide theoretical foundation for LoD's viability, and extensive experiments verify its superiority.

Figures

Figures reproduced from arXiv: 2505.12952 by the authors.

**Figure 1.** The cross-entropy loss changes of ID (label-noise) and OOD (label-clean) samples in unlabeled wild data when they are intentionally labeled as K + 1-th class. These two types of samples typically exhibit different loss curves due to the differences in how learning progresses for each. samples in unlabeled wild data is intentionally transformed into label-clean samples, while the ID counterparts become label-noise on… view at source ↗

**Figure 2.** Overview of the loss-difference OOD detection framework by intentionally label-noisifying unlabeled wild data. [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗

**Figure 3.** The mean cross-entropy loss curves respectively for all ID [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗

Figures from the paper (2 more)

**Figure 4.** Figure 4: Experiments in different rations (|Btrain in |/|Bwild|) on standard benchmarks (dashed lines) and hard benchmarks (solid lines). the results. As the ratio increases, the model performance consistently improves across all benchmarks, strongly supporting our claim. Con…

**Figure 5.** Figure 5: The impacts of training epochs on results respectively in [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

45 extracted references · 31 canonical work pages

[1]

Latent space autore- gression for novelty detection

[Abatiet al., 2019 ] Davide Abati, Angelo Porrello, Simone Calderara, and Rita Cucchiara. Latent space autore- gression for novelty detection. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 481–490,

work page 2019
[10]

How does unlabeled data prov- ably help out-of-distribution detection?arXiv preprint arXiv:2402.03502,

[Duet al., 2024 ] Xuefeng Du, Zhen Fang, Ilias Diakoniko- las, and Yixuan Li. How does unlabeled data prov- ably help out-of-distribution detection?arXiv preprint arXiv:2402.03502,

arXiv 2024
[11]

On the learnability of out-of-distribution de- tection.Journal of Machine Learning Research, 25,

[Fanget al., 2024 ] Zhen Fang, Yixuan Li, Feng Liu, Bo Han, and Jie Lu. On the learnability of out-of-distribution de- tection.Journal of Machine Learning Research, 25,

work page 2024
[12]

Leveraging Unlabeled Data to Track Memorization

[Forouzeshet al., 2022 ] Mahsa Forouzesh, Hanie Sedghi, and Patrick Thiran. Leveraging unlabeled data to track memorization.arXiv preprint arXiv:2212.04461,

work page Pith review arXiv 2022
[13]

Who said what: Modeling indi- vidual labelers improves classification

[Guanet al., 2018 ] Melody Guan, Varun Gulshan, Andrew Dai, and Geoffrey Hinton. Who said what: Modeling indi- vidual labelers improves classification. InProceedings of the AAAI conference on artificial intelligence, volume 32,

work page 2018
[14]

A baseline for detecting misclassified and out-of- distribution examples in neural networks.arXiv preprint arXiv:1610.02136,

[Hendrycks and Gimpel, 2016] Dan Hendrycks and Kevin Gimpel. A baseline for detecting misclassified and out-of- distribution examples in neural networks.arXiv preprint arXiv:1610.02136,

arXiv 2016
[16]

Training ood de- tectors in their natural habitats

[Katz-Samuelset al., 2022 ] Julian Katz-Samuels, Julia B Nakhleh, Robert Nowak, and Yixuan Li. Training ood de- tectors in their natural habitats. InInternational Confer- ence on Machine Learning, pages 10848–10865. PMLR,

work page 2022
[17]

Learning multiple lay- ers of features from tiny images

[Krizhevsky, 2009] Alex Krizhevsky. Learning multiple lay- ers of features from tiny images

work page 2009

Show all 45 references

[18]

Training confidence-calibrated classifiers for detecting out-of-distribution samples.arXiv preprint arXiv:1711.09325,

[Leeet al., 2017 ] Kimin Lee, Honglak Lee, Kibok Lee, and Jinwoo Shin. Training confidence-calibrated classifiers for detecting out-of-distribution samples.arXiv preprint arXiv:1711.09325,

2017 arXiv
[19]

A simple unified framework for detect- ing out-of-distribution samples and adversarial attacks

[Leeet al., 2018 ] Kimin Lee, Kibok Lee, Honglak Lee, and Jinwoo Shin. A simple unified framework for detect- ing out-of-distribution samples and adversarial attacks. Advances in neural information processing systems, 31,

2018
[20]

Disc: Learning from noisy labels via dynamic instance-specific selection and correction

[Liet al., 2023 ] Yifan Li, Hu Han, Shiguang Shan, and Xilin Chen. Disc: Learning from noisy labels via dynamic instance-specific selection and correction. InProceedings of the IEEE/CVF Conference on Computer Vision and Pat- tern Recognition, pages 24070–24079,

2023
[21]

Enhancing the reliability of out-of-distribution image detection in neural networks

[Lianget al., 2017 ] Shiyu Liang, Yixuan Li, and Rayadurgam Srikant. Enhancing the reliability of out-of-distribution image detection in neural networks. arXiv preprint arXiv:1706.02690,

2017 arXiv
[22]

Mitigating label noise through data ambigua- tion

[Lienen and Hüllermeier, 2024] Julian Lienen and Eyke Hüllermeier. Mitigating label noise through data ambigua- tion. InProceedings of the AAAI Conference on Artificial Intelligence, volume 38, pages 13799–13807,

2024
[23]

Learning the latent causal structure for modeling label noise

[Linet al., 2024 ] Yexiong Lin, Yu Yao, and Tongliang Liu. Learning the latent causal structure for modeling label noise. InThe Thirty-eighth Annual Conference on Neu- ral Information Processing Systems,

2024
[25]

Open set learning with counterfactual images

[Nealet al., 2018 ] Lawrence Neal, Matthew Olson, Xiaoli Fern, Weng-Keen Wong, and Fuxin Li. Open set learning with counterfactual images. InProceedings of the Euro- pean Conference on Computer Vision (ECCV), pages 613– 628,

2018
[26]

Reading digits in natural images with unsupervised feature learning

[Netzeret al., 2011 ] Yuval Netzer, Tao Wang, Adam Coates, Alessandro Bissacco, Baolin Wu, Andrew Y Ng, et al. Reading digits in natural images with unsupervised feature learning. InNIPS workshop on deep learning and unsu- pervised feature learning, volume 2011, page

2011
[30]

Dice: Lever- aging sparsification for out-of-distribution detection

[Sun and Li, 2022] Yiyou Sun and Yixuan Li. Dice: Lever- aging sparsification for out-of-distribution detection. In European Conference on Computer Vision, pages 691–

2022
[31]

React: Out-of-distribution detection with rectified activa- tions.Advances in Neural Information Processing Sys- tems, 34:144–157,

[Sunet al., 2021 ] Yiyou Sun, Chuan Guo, and Yixuan Li. React: Out-of-distribution detection with rectified activa- tions.Advances in Neural Information Processing Sys- tems, 34:144–157,

2021
[32]

Out-of-distribution detection with deep near- est neighbors

[Sunet al., 2022 ] Yiyou Sun, Yifei Ming, Xiaojin Zhu, and Yixuan Li. Out-of-distribution detection with deep near- est neighbors. InInternational Conference on Machine Learning, pages 20827–20840. PMLR,

2022
[33]

Csi: Novelty detection via contrastive learning on distributionally shifted in- stances.Advances in neural information processing sys- tems, 33:11839–11852,

[Tacket al., 2020 ] Jihoon Tack, Sangwoo Mo, Jongheon Jeong, and Jinwoo Shin. Csi: Novelty detection via contrastive learning on distributionally shifted in- stances.Advances in neural information processing sys- tems, 33:11839–11852,

2020
[34]

Open-set recognition: A good closed-set classifier is all you need.the International Con- ference on Learning Representations, abs/2110.06207,

[Vazeet al., 2022 ] Sagar Vaze, Kai Han, Andrea Vedaldi, and Andrew Zisserman. Open-set recognition: A good closed-set classifier is all you need.the International Con- ference on Learning Representations, abs/2110.06207,

2022 arXiv
[35]

Vim: Out-of-distribution with virtual- logit matching

[Wanget al., 2022 ] Haoqi Wang, Zhizhong Li, Litong Feng, and Wayne Zhang. Vim: Out-of-distribution with virtual- logit matching. InProceedings of the IEEE/CVF confer- ence on computer vision and pattern recognition, pages 4921–4930,

2022
[37]

Generalized out-of-distribution detec- tion: A survey.International Journal of Computer Vision, 132(12):5635–5662,

[Yanget al., 2024 ] Jingkang Yang, Kaiyang Zhou, Yixuan Li, and Ziwei Liu. Generalized out-of-distribution detec- tion: A survey.International Journal of Computer Vision, 132(12):5635–5662,

2024
[38]

Lsun: Construction of a large-scale image dataset using deep learning with humans in the loop.arXiv preprint arXiv:1506.03365,

[Yuet al., 2015 ] Fisher Yu, Ari Seff, Yinda Zhang, Shu- ran Song, Thomas Funkhouser, and Jianxiong Xiao. Lsun: Construction of a large-scale image dataset using deep learning with humans in the loop.arXiv preprint arXiv:1506.03365,

2015 arXiv
[39]

Ctrl: Clus- tering training losses for label error detection.IEEE Trans- actions on Artificial Intelligence,

[Yue and Jha, 2024] Chang Yue and Niraj K Jha. Ctrl: Clus- tering training losses for label error detection.IEEE Trans- actions on Artificial Intelligence,

2024
[40]

Wide residual net- works.arXiv preprint arXiv:1605.07146,

[Zagoruyko, 2016] Sergey Zagoruyko. Wide residual net- works.arXiv preprint arXiv:1605.07146,

2016 arXiv
[41]

Out-of-distribution detection learning with unreliable out- of-distribution sources.Advances in Neural Information Processing Systems, 36:72110–72123,

[Zhenget al., 2023 ] Haotian Zheng, Qizhou Wang, Zhen Fang, Xiaobo Xia, Feng Liu, Tongliang Liu, and Bo Han. Out-of-distribution detection learning with unreliable out- of-distribution sources.Advances in Neural Information Processing Systems, 36:72110–72123,

2023
[42]

Places: A 10 million image database for scene recognition.IEEE transactions on pattern analysis and machine intelligence, 40(6):1452–1464,

[Zhouet al., 2017 ] Bolei Zhou, Agata Lapedriza, Aditya Khosla, Aude Oliva, and Antonio Torralba. Places: A 10 million image database for scene recognition.IEEE transactions on pattern analysis and machine intelligence, 40(6):1452–1464,

2017
[43]

Diversified outlier exposure for out- of-distribution detection via informative extrapolation

[Zhuet al., 2023 ] Jianing Zhu, Yu Geng, Jiangchao Yao, Tongliang Liu, Gang Niu, Masashi Sugiyama, and Bo Han. Diversified outlier exposure for out- of-distribution detection via informative extrapolation. Advances in Neural Information Processing Systems, 36:22702–22734,

2023
[44]

A Detailed Proof To prove Proposition 1, we first reintroduce Lemma 1 from [Liuet al., 2020a ] and Proposition 1 as follows: Lemma 1(Early-learning succeeds).Denote by {θ t} the it- erates of gradient descent with step sizeη. For any∆∈ (0,1/2), there exists a constantδ ∆, depe...

2024
[45]

Letx=v+z i, wherezi∼N(0,σ 2Id×d)

=sig(θ Txi) = 1 1 +e−θTxi , p(yi =−1) = 1−p(y i = 1). Letx=v+z i, wherezi∼N(0,σ 2Id×d). For each sample xi∈D wild out (label-clean), we uselogfor its loss, and have li(θ) = log(1 +e−θT (v+zi))≤e −θT (v+zi). Similarly, for each samplexj∈D wild in (label-noise), we have lj(θ) = ...
[2009]

Extremely simple ac- tivation shaping for out-of-distribution detection.arXiv preprint arXiv:2209.09858,

[Djurisicet al., 2022 ] Andrija Djurisic, Nebojsa Bozanic, Arjun Ashok, and Rosanne Liu. Extremely simple ac- tivation shaping for out-of-distribution detection.arXiv preprint arXiv:2209.09858,

2022 arXiv
[2011]

Deep neural networks are easily fooled: High con- fidence predictions for unrecognizable images

[Nguyenet al., 2015 ] Anh Nguyen, Jason Yosinski, and Jeff Clune. Deep neural networks are easily fooled: High con- fidence predictions for unrecognizable images. InPro- ceedings of the IEEE conference on computer vision and pattern recognition, pages 427–436,

2015
[2014]

Imagenet: A large-scale hierarchical image database

[Denget al., 2009 ] Jia Deng, Wei Dong, Richard Socher, Li- Jia Li, Kai Li, and Li Fei-Fei. Imagenet: A large-scale hierarchical image database. In2009 IEEE conference on computer vision and pattern recognition, pages 248–255. Ieee,

2009
[2015]

Gradient- regularized out-of-distribution detection

[Sharifiet al., 2025 ] Sina Sharifi, Taha Entesari, Bardia Safaei, Vishal M Patel, and Mahyar Fazlyab. Gradient- regularized out-of-distribution detection. InEuropean Conference on Computer Vision, pages 459–478. Springer,

2025
[2016]

Deep anomaly detection with out- lier exposure.arXiv preprint arXiv:1812.04606,

[Hendryckset al., 2018 ] Dan Hendrycks, Mantas Mazeika, and Thomas Dietterich. Deep anomaly detection with out- lier exposure.arXiv preprint arXiv:1812.04606,

2018 arXiv
[2017]

Gradorth: a simple yet efficient out-of-distribution detection with or- thogonal projection of gradients.Advances in Neural In- formation Processing Systems, 36,

[Behpouret al., 2024 ] Sima Behpour, Thang Long Doan, Xin Li, Wenbin He, Liang Gou, and Liu Ren. Gradorth: a simple yet efficient out-of-distribution detection with or- thogonal projection of gradients.Advances in Neural In- formation Processing Systems, 36,

2024
[2018]

Atom: Robustifying out-of- distribution detection using outlier mining

[Chenet al., 2021 ] Jiefeng Chen, Yixuan Li, Xi Wu, Yingyu Liang, and Somesh Jha. Atom: Robustifying out-of- distribution detection using outlier mining. InMachine Learning and Knowledge Discovery in Databases. Re- search Track: European Conference, ECML PKDD 2021, Bilbao, Spa...

2021
[2019]

A closer look at memo- rization in deep networks

[Arpitet al., 2017 ] Devansh Arpit, Stanisław Jastrz˛ ebski, Nicolas Ballas, David Krueger, Emmanuel Bengio, Maxinder S Kanwal, Tegan Maharaj, Asja Fischer, Aaron Courville, Yoshua Bengio, et al. A closer look at memo- rization in deep networks. InInternational conference on m...

2017
[2020]

Predictive uncertainty estimation via prior networks.Ad- vances in neural information processing systems, 31,

[Malinin and Gales, 2018] Andrey Malinin and Mark Gales. Predictive uncertainty estimation via prior networks.Ad- vances in neural information processing systems, 31,

2018
[2021]

Describing textures in the wild

[Cimpoiet al., 2014 ] Mircea Cimpoi, Subhransu Maji, Ia- sonas Kokkinos, Sammy Mohamed, and Andrea Vedaldi. Describing textures in the wild. InProceedings of the IEEE conference on computer vision and pattern recog- nition, pages 3606–3613,

2014
[2022]

Unknown-aware object detection: Learn- ing what you don’t know from videos in the wild

[Duet al., 2022 ] Xuefeng Du, Xin Wang, Gabriel Gozum, and Yixuan Li. Unknown-aware object detection: Learn- ing what you don’t know from videos in the wild. InPro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 13678–13688,

2022
[2023]

Mitigating neural net- work overconfidence with logit normalization

[Weiet al., 2022 ] Hongxin Wei, Renchunzi Xie, Hao Cheng, Lei Feng, Bo An, and Yixuan Li. Mitigating neural net- work overconfidence with logit normalization. InInter- national conference on machine learning, pages 23631– 23644. PMLR,

2022
[2024]

Discriminative out-of- distribution detection for semantic segmentation.arXiv preprint arXiv:1808.07703,

[Bevandi´cet al., 2018 ] Petra Bevandi ´c, Ivan Krešo, Marin Orši´c, and Siniša Šegvi ´c. Discriminative out-of- distribution detection for semantic segmentation.arXiv preprint arXiv:1808.07703,

2018 arXiv
[2025]

Learning from noisy labels with deep neural networks: A survey.IEEE transactions on neural networks and learning systems, 34(11):8135–8153,

[Songet al., 2022 ] Hwanjun Song, Minseok Kim, Dong- min Park, Yooju Shin, and Jae-Gil Lee. Learning from noisy labels with deep neural networks: A survey.IEEE transactions on neural networks and learning systems, 34(11):8135–8153,

2022

Pith tools

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

[1] [1]

Latent space autore- gression for novelty detection

[Abatiet al., 2019 ] Davide Abati, Angelo Porrello, Simone Calderara, and Rita Cucchiara. Latent space autore- gression for novelty detection. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 481–490,

work page 2019

[2] [10]

How does unlabeled data prov- ably help out-of-distribution detection?arXiv preprint arXiv:2402.03502,

[Duet al., 2024 ] Xuefeng Du, Zhen Fang, Ilias Diakoniko- las, and Yixuan Li. How does unlabeled data prov- ably help out-of-distribution detection?arXiv preprint arXiv:2402.03502,

arXiv 2024

[3] [11]

On the learnability of out-of-distribution de- tection.Journal of Machine Learning Research, 25,

[Fanget al., 2024 ] Zhen Fang, Yixuan Li, Feng Liu, Bo Han, and Jie Lu. On the learnability of out-of-distribution de- tection.Journal of Machine Learning Research, 25,

work page 2024

[4] [12]

Leveraging Unlabeled Data to Track Memorization

[Forouzeshet al., 2022 ] Mahsa Forouzesh, Hanie Sedghi, and Patrick Thiran. Leveraging unlabeled data to track memorization.arXiv preprint arXiv:2212.04461,

work page Pith review arXiv 2022

[5] [13]

Who said what: Modeling indi- vidual labelers improves classification

[Guanet al., 2018 ] Melody Guan, Varun Gulshan, Andrew Dai, and Geoffrey Hinton. Who said what: Modeling indi- vidual labelers improves classification. InProceedings of the AAAI conference on artificial intelligence, volume 32,

work page 2018

[6] [14]

A baseline for detecting misclassified and out-of- distribution examples in neural networks.arXiv preprint arXiv:1610.02136,

[Hendrycks and Gimpel, 2016] Dan Hendrycks and Kevin Gimpel. A baseline for detecting misclassified and out-of- distribution examples in neural networks.arXiv preprint arXiv:1610.02136,

arXiv 2016

[7] [16]

Training ood de- tectors in their natural habitats

[Katz-Samuelset al., 2022 ] Julian Katz-Samuels, Julia B Nakhleh, Robert Nowak, and Yixuan Li. Training ood de- tectors in their natural habitats. InInternational Confer- ence on Machine Learning, pages 10848–10865. PMLR,

work page 2022

[8] [17]

Learning multiple lay- ers of features from tiny images

[Krizhevsky, 2009] Alex Krizhevsky. Learning multiple lay- ers of features from tiny images

work page 2009