pith. sign in

arxiv: 2605.23797 · v1 · pith:I64M7QVHnew · submitted 2026-05-22 · 💻 cs.LG · cs.CV

Debiased Negative Mining Improves Out-of-distribution Detection with Pre-trained Vision-Language Models

Bo Peng , Jie Lu , Guangquan Zhang , Zhen Fang This is my paper

Pith reviewed 2026-05-25 04:47 UTC · model grok-4.3

classification 💻 cs.LG cs.CV
keywords out-of-distribution detectionvision-language modelsnegative label miningdebiased samplingMonte-Carlo samplingfalse negative problempost-hoc detection
0
0 comments X

The pith

Debiased negative mining converts to Monte-Carlo sampling from ID labels and wild data to improve VLM-based OOD detection.

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

The paper develops a theoretical framework that corrects sampling bias when mining negative labels from unlabeled wild corpus data for out-of-distribution detection with pre-trained vision-language models. Current heuristic approaches suffer from false negatives that limit how well affinities between inputs and labels can separate in-distribution from out-of-distribution cases. The framework shows that debiased mining reduces to straightforward Monte-Carlo sampling using only ID labels and the same wild data. This change yields stronger OOD scoring without access to target OOD labels. A reader would care because more reliable detection of unexpected inputs directly improves the safety of deployed machine learning systems.

Core claim

The central claim is that a theoretical framework for correcting the sampling bias of negative labels by indirectly approximating their distribution allows the debiased negative mining procedure to be converted into Monte-Carlo sampling based on ID labels and the unlabeled wild corpus data, which establishes a new state-of-the-art across a variety of OOD detection setups that use pre-trained vision-language models.

What carries the argument

Theoretical framework that corrects sampling bias of negative labels by indirectly approximating their distribution and thereby converts debiased mining into Monte-Carlo sampling.

If this is right

  • The method achieves new state-of-the-art results in multiple OOD detection setups that rely on pre-trained vision-language models.
  • It directly mitigates the false negative problem that arises when mining negative labels from unlabeled wild corpus data.
  • It operates using only ID labels and the unlabeled wild corpus without requiring any target OOD labels.
  • It improves post-hoc OOD scoring that examines affinities between inputs and both ID and negative labels.

Where Pith is reading between the lines

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

  • The Monte-Carlo formulation could simplify deployment in settings where wild unlabeled data is abundant but labeled OOD examples are scarce.
  • Similar bias-correction steps might transfer to other label-mining tasks that use pre-trained models for anomaly or novelty detection.
  • Testing the approach on non-vision modalities or different VLM architectures would check whether the sampling conversion remains effective.
  • The framework may connect to existing sampling techniques used in semi-supervised learning or active learning.

Load-bearing premise

The sampling bias of negative labels can be corrected by indirectly approximating the distribution of negative labels via the proposed theoretical framework.

What would settle it

Apply the Monte-Carlo sampling version to standard OOD benchmarks and measure whether it outperforms heuristic negative mining; failure to improve or to reduce the false-negative rate would falsify the central claim.

Figures

Figures reproduced from arXiv: 2605.23797 by Bo Peng, Guangquan Zhang, Jie Lu, Zhen Fang.

Figure 1
Figure 1. Figure 1: Hyper-parameter analysis on ImageNet-1K w.r.t. [PITH_FULL_IMAGE:figures/full_fig_p008_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Hyper-parameter analysis results on ImageNet-1K w.r.t. [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗
read the original abstract

Aiming at identifying unexpected inputs from unknown classes, out-of-distribution (OOD) detection has emerged as a pivotal approach to enhancing the reliability of machine learning models. This paper focuses on the burgeoning paradigm of post-hoc OOD detection with pre-trained vision-language models (VLMs), where a popular pipeline is to detect OOD inputs by examining their affinities between ID labels and negative labels, i.e., those semantically different from ID labels. Due to the unavailability of target OOD labels, existing works predominantly rely on heuristic rules to mine negative labels from unlabeled wild corpus data. Despite the empirical success, we argue that the power of VLM-based OOD detection has yet to be fully unleashed since the notorious false negative problem is far from addressed in the literature. With this motivation, we are interested in addressing the challenge of mining true negative labels for OOD scoring. To this end, we develop a theoretical framework for correcting the sampling bias of negatives labels by indirectly approximating the distribution of negative labels. Perhaps surprisingly, we show that the debiased negative mining can be naturally converted into Monte-Carlo sampling based on ID labels and the unlabeled wild corpus data. Extensive experiments empirically manifest that our method establishes a new state-of-the-art in a variety of OOD detection setups. Code is publicly available at \href{https://github.com/60pen9/Debiased-Negative-Mining-Improves-OOD-Detection-with-Pre-trained-VLMs}{\textcolor{red}{here}}.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

1 major / 2 minor

Summary. The paper proposes a theoretical framework to correct sampling bias when mining negative labels from unlabeled wild corpus data for post-hoc OOD detection with pre-trained VLMs. It claims that this debiased approach converts naturally into Monte-Carlo sampling using only ID labels and the wild corpus, and reports new state-of-the-art empirical results across multiple OOD detection setups.

Significance. If the central theoretical approximation holds, the work could strengthen VLM-based OOD detection by addressing false negatives in negative mining, with relevance to reliability in deployed systems. The conversion to observable-data Monte-Carlo sampling (if rigorously established) and the public code release are clear strengths supporting reproducibility.

major comments (1)
  1. [theoretical framework (abstract and main derivation)] Abstract and theoretical framework section: the claim that debiased negative mining 'can be naturally converted into Monte-Carlo sampling' rests on an indirect approximation of the negative-label distribution. No explicit bound, convergence argument, or factorization assumption verification is supplied to guarantee that the resulting estimator remains unbiased when the wild corpus contains residual semantic overlap with ID classes or when VLM similarities exhibit heavy tails.
minor comments (2)
  1. [Abstract] Abstract: 'empirically manifest' is nonstandard phrasing; 'demonstrate' or 'show' would be clearer.
  2. [theoretical framework] The manuscript should clarify the precise conditions under which the Monte-Carlo estimator is unbiased, ideally with a short lemma or proposition.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the careful review and constructive feedback on the theoretical framework. We address the major comment below.

read point-by-point responses

  1. Referee: [theoretical framework (abstract and main derivation)] Abstract and theoretical framework section: the claim that debiased negative mining 'can be naturally converted into Monte-Carlo sampling' rests on an indirect approximation of the negative-label distribution. No explicit bound, convergence argument, or factorization assumption verification is supplied to guarantee that the resulting estimator remains unbiased when the wild corpus contains residual semantic overlap with ID classes or when VLM similarities exhibit heavy tails.

    Authors: The derivation begins from the target debiased distribution over true negative labels and rewrites the relevant expectation by conditioning on the observable wild corpus and ID labels, yielding the Monte-Carlo estimator via the law of total expectation. The step relies on the modeling assumption that negatives are those with low VLM similarity to any ID class. We acknowledge that the manuscript supplies neither explicit error bounds on the approximation nor a verification of the factorization under residual semantic overlap or heavy-tailed similarities; these are genuine limitations of the current presentation. We will revise the theoretical framework section to include an expanded discussion of the modeling assumptions, the conditions under which the estimator remains approximately unbiased, and the potential bias introduced by overlap or heavy tails. revision: yes

Circularity Check

0 steps flagged

No circularity: derivation rests on independent theoretical approximation

full rationale

The paper presents a theoretical framework that corrects sampling bias via indirect approximation of negative label distribution, then converts the debiased mining into Monte-Carlo sampling using ID labels and wild corpus data. No equations, self-citations, or fitted parameters are shown reducing the central claim to its inputs by construction. The conversion is framed as a derived consequence of the framework rather than a renaming or self-referential fit, leaving the method self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

Based solely on the abstract, the central claim rests on the assumption that negative label distribution can be indirectly approximated; no explicit free parameters or invented entities are mentioned.

axioms (1)
  • domain assumption The distribution of negative labels can be indirectly approximated from ID labels and unlabeled wild corpus data to correct sampling bias.
    This is the core premise of the theoretical framework described in the abstract for enabling the Monte-Carlo conversion.

pith-pipeline@v0.9.0 · 5804 in / 1285 out tokens · 31185 ms · 2026-05-25T04:47:30.757829+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

75 extracted references · 75 canonical work pages · 4 internal anchors

  1. [1]

    Differentiated graph regularized non-negative matrix factor- ization for semi-supervised community detection.Physica A: Statistical Mechanics and its Applications, 604:127692, 2022

    Chunchun Chen, Wenjie Zhu, and Bo Peng. Differentiated graph regularized non-negative matrix factor- ization for semi-supervised community detection.Physica A: Statistical Mechanics and its Applications, 604:127692, 2022

    work page 2022

  2. [2]

    Towards robust community detection via extreme adversarial attacks

    Chunchun Chen, Wenjie Zhu, Bo Peng, and Huijuan Lu. Towards robust community detection via extreme adversarial attacks. In2022 26th International Conference on Pattern Recognition (ICPR), pp. 2231–

    work page

  3. [3]

    Conjugated semantic pool improves ood detection with pre-trained vision-language models.Advances in Neural Information Processing Systems, 37:82560– 82593, 2024

    Mengyuan Chen, Junyu Gao, and Changsheng Xu. Conjugated semantic pool improves ood detection with pre-trained vision-language models.Advances in Neural Information Processing Systems, 37:82560– 82593, 2024

    work page 2024

  4. [4]

    Understanding transferable representation learning and zero-shot transfer in clip.arXiv preprint arXiv:2310.00927, 2023

    Zixiang Chen, Yihe Deng, Yuanzhi Li, and Quanquan Gu. Understanding transferable representation learning and zero-shot transfer in clip.arXiv preprint arXiv:2310.00927, 2023

    work page arXiv 2023

  5. [5]

    Describing textures in the wild

    Mircea Cimpoi, Subhransu Maji, Iasonas Kokkinos, Sammy Mohamed, and Andrea Vedaldi. Describing textures in the wild. InProceedings of the IEEE conference on computer vision and pattern recognition, pp. 3606–3613, 2014

    work page 2014

  6. [6]

    Imagenet: A large-scale hierarchi- cal image database

    Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. Imagenet: A large-scale hierarchi- cal image database. In2009 IEEE conference on computer vision and pattern recognition, pp. 248–255. Ieee, 2009

    work page 2009

  7. [7]

    Reducing network agnostophobia.Advances in Neural Information Processing Systems, 31, 2018

    Akshay Raj Dhamija, Manuel G ¨unther, and Terrance Boult. Reducing network agnostophobia.Advances in Neural Information Processing Systems, 31, 2018

    work page 2018

  8. [8]

    Towards unknown-aware learning with virtual outlier synthesis

    Xuefeng Du, Zhaoning Wang, Mu Cai, and Sharon Li. Towards unknown-aware learning with virtual outlier synthesis. InInternational Conference on Learning Representations, volume 1, pp. 5, 2022

    work page 2022

  9. [9]

    Vos: Learning what you don't know by virtual outlier synthesis

    Xuefeng Du, Zhaoning Wang, Mu Cai, and Yixuan Li. V os: Learning what you don’t know by virtual outlier synthesis.arXiv preprint arXiv:2202.01197, 2022

    work page arXiv 2022

  10. [10]

    How does wild data provably help ood detection? InThe Twelfth International Conference on Learning Representations, 2024

    Xuefeng Du, Zhen Fang, Ilias Diakonikolas, and Yixuan Li. How does wild data provably help ood detection? InThe Twelfth International Conference on Learning Representations, 2024

    work page 2024

  11. [11]

    Convex formulation for learning from positive and unlabeled data

    Marthinus Du Plessis, Gang Niu, and Masashi Sugiyama. Convex formulation for learning from positive and unlabeled data. InInternational conference on machine learning, pp. 1386–1394. PMLR, 2015. 11

    work page 2015

  12. [12]

    Analysis of learning from positive and unlabeled data.Advances in neural information processing systems, 27, 2014

    Marthinus C Du Plessis, Gang Niu, and Masashi Sugiyama. Analysis of learning from positive and unlabeled data.Advances in neural information processing systems, 27, 2014

    work page 2014

  13. [13]

    Learning classifiers from only positive and unlabeled data

    Charles Elkan and Keith Noto. Learning classifiers from only positive and unlabeled data. InProceedings of the 14th ACM SIGKDD international conference on Knowledge discovery and data mining, pp. 213– 220, 2008

    work page 2008

  14. [14]

    Zero-shot out-of-distribution detection based on the pre-trained model clip

    Sepideh Esmaeilpour, Bing Liu, Eric Robertson, and Lei Shu. Zero-shot out-of-distribution detection based on the pre-trained model clip. InProceedings of the AAAI conference on artificial intelligence, volume 36, pp. 6568–6576, 2022

    work page 2022

  15. [15]

    A Baseline for Detecting Misclassified and Out-of-Distribution Examples in Neural Networks

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

    work page internal anchor Pith review Pith/arXiv arXiv 2016

  16. [16]

    Deep Anomaly Detection with Outlier Exposure

    Dan Hendrycks, Mantas Mazeika, and Thomas Dietterich. Deep anomaly detection with outlier exposure. arXiv preprint arXiv:1812.04606, 2018

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  17. [17]

    Scaling out-of-distribution detection for real-world settings

    Dan Hendrycks, Steven Basart, Mantas Mazeika, Andy Zou, Joe Kwon, Mohammadreza Mostajabi, Jacob Steinhardt, and Dawn Song. Scaling out-of-distribution detection for real-world settings.arXiv preprint arXiv:1911.11132, 2019

    work page arXiv 1911

  18. [18]

    The many faces of robustness: A critical analysis of out- of-distribution generalization

    Dan Hendrycks, Steven Basart, Norman Mu, Saurav Kadavath, Frank Wang, Evan Dorundo, Rahul Desai, Tyler Zhu, Samyak Parajuli, Mike Guo, et al. The many faces of robustness: A critical analysis of out- of-distribution generalization. InProceedings of the IEEE/CVF international conference on computer vision, pp. 8340–8349, 2021

    work page 2021

  19. [19]

    Natural adversarial examples

    Dan Hendrycks, Kevin Zhao, Steven Basart, Jacob Steinhardt, and Dawn Song. Natural adversarial examples. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp. 15262–15271, 2021

    work page 2021

  20. [20]

    On the importance of gradients for detecting distributional shifts in the wild.Advances in Neural Information Processing Systems, 34:677–689, 2021

    Rui Huang, Andrew Geng, and Yixuan Li. On the importance of gradients for detecting distributional shifts in the wild.Advances in Neural Information Processing Systems, 34:677–689, 2021

    work page 2021

  21. [21]

    A survey of safety and trustworthiness of deep neural networks: Verification, testing, adversarial attack and defence, and interpretability.Computer Science Review, 37:100270, 2020

    Xiaowei Huang, Daniel Kroening, Wenjie Ruan, James Sharp, Youcheng Sun, Emese Thamo, Min Wu, and Xinping Yi. A survey of safety and trustworthiness of deep neural networks: Verification, testing, adversarial attack and defence, and interpretability.Computer Science Review, 37:100270, 2020

    work page 2020

  22. [22]

    Learning representation for clustering via prototype scattering and positive sampling.IEEE Transactions on Pattern Analysis and Machine Intelligence, 45(6):7509–7524, 2022

    Zhizhong Huang, Jie Chen, Junping Zhang, and Hongming Shan. Learning representation for clustering via prototype scattering and positive sampling.IEEE Transactions on Pattern Analysis and Machine Intelligence, 45(6):7509–7524, 2022

    work page 2022

  23. [23]

    Robust estimation of a location parameter

    Peter J Huber. Robust estimation of a location parameter. InBreakthroughs in statistics: Methodology and distribution, pp. 492–518. Springer, 1992

    work page 1992

  24. [24]

    Negative label guided ood detection with pretrained vision-language models.arXiv preprint arXiv:2403.20078, 2024

    Xue Jiang, Feng Liu, Zhen Fang, Hong Chen, Tongliang Liu, Feng Zheng, and Bo Han. Negative label guided ood detection with pretrained vision-language models.arXiv preprint arXiv:2403.20078, 2024

    work page arXiv 2024

  25. [25]

    Training ood detectors in their natural habitats

    Julian Katz-Samuels, Julia B Nakhleh, Robert Nowak, and Yixuan Li. Training ood detectors in their natural habitats. InInternational Conference on Machine Learning, pp. 10848–10865. PMLR, 2022

    work page 2022

  26. [26]

    Being bayesian, even just a bit, fixes overconfi- dence in relu networks

    Agustinus Kristiadi, Matthias Hein, and Philipp Hennig. Being bayesian, even just a bit, fixes overconfi- dence in relu networks. InInternational conference on machine learning, pp. 5436–5446. PMLR, 2020

    work page 2020

  27. [27]

    Training Confidence-calibrated Classifiers for Detecting Out-of-Distribution Samples

    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

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  28. [28]

    A simple unified framework for detecting out- of-distribution samples and adversarial attacks.Advances in neural information processing systems, 31, 2018

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

    work page 2018

  29. [29]

    Learning transferable negative prompts for out-of-distribution detection

    Tianqi Li, Guansong Pang, Xiao Bai, Wenjun Miao, and Jin Zheng. Learning transferable negative prompts for out-of-distribution detection. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 17584–17594, 2024

    work page 2024

  30. [30]

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

    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

    work page arXiv 2017

  31. [31]

    Estimating the partition function by discrim- inance sampling

    Qiang Liu, Jian Peng, Alexander Ihler, and John Fisher III. Estimating the partition function by discrim- inance sampling. InProceedings of the Thirty-First Conference on Uncertainty in Artificial Intelligence, pp. 514–522, 2015. 12

    work page 2015

  32. [32]

    Energy-based out-of-distribution detection

    Weitang Liu, Xiaoyun Wang, John Owens, and Yixuan Li. Energy-based out-of-distribution detection. Advances in neural information processing systems, 33:21464–21475, 2020

    work page 2020

  33. [33]

    Predictive uncertainty estimation via prior networks.Advances in neural information processing systems, 31, 2018

    Andrey Malinin and Mark Gales. Predictive uncertainty estimation via prior networks.Advances in neural information processing systems, 31, 2018

    work page 2018

  34. [34]

    Reverse kl-divergence training of prior networks: Improved uncertainty and adversarial robustness.Advances in Neural Information Processing Systems, 32, 2019

    Andrey Malinin and Mark Gales. Reverse kl-divergence training of prior networks: Improved uncertainty and adversarial robustness.Advances in Neural Information Processing Systems, 32, 2019

    work page 2019

  35. [35]

    Wordnet: a lexical database for english.Communications of the ACM, 38(11):39–41, 1995

    George A Miller. Wordnet: a lexical database for english.Communications of the ACM, 38(11):39–41, 1995

    work page 1995

  36. [36]

    Delving into out-of-distribution detection with vision-language representations.Advances in neural information processing systems, 35: 35087–35102, 2022

    Yifei Ming, Ziyang Cai, Jiuxiang Gu, Yiyou Sun, Wei Li, and Yixuan Li. Delving into out-of-distribution detection with vision-language representations.Advances in neural information processing systems, 35: 35087–35102, 2022

    work page 2022

  37. [37]

    Locoop: Few-shot out-of-distribution detection via prompt learning.Advances in Neural Information Processing Systems, 36, 2024

    Atsuyuki Miyai, Qing Yu, Go Irie, and Kiyoharu Aizawa. Locoop: Few-shot out-of-distribution detection via prompt learning.Advances in Neural Information Processing Systems, 36, 2024

    work page 2024

  38. [38]

    Provable guarantees for understanding out-of-distribution detection

    Peyman Morteza and Yixuan Li. Provable guarantees for understanding out-of-distribution detection. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 36, pp. 7831–7840, 2022

    work page 2022

  39. [39]

    Out of distribution data detection using dropout bayesian neural networks

    Andre T Nguyen, Fred Lu, Gary Lopez Munoz, Edward Raff, Charles Nicholas, and James Holt. Out of distribution data detection using dropout bayesian neural networks. InProceedings of the AAAI Confer- ence on Artificial Intelligence, volume 36, pp. 7877–7885, 2022

    work page 2022

  40. [40]

    Deep neural networks are easily fooled: High confidence predictions for unrecognizable images

    Anh Nguyen, Jason Yosinski, and Jeff Clune. Deep neural networks are easily fooled: High confidence predictions for unrecognizable images. InProceedings of the IEEE conference on computer vision and pattern recognition, pp. 427–436, 2015

    work page 2015

  41. [41]

    Out-of-distribution detection with negative prompts

    Jun Nie, Yonggang Zhang, Zhen Fang, Tongliang Liu, Bo Han, and Xinmei Tian. Out-of-distribution detection with negative prompts. InThe Twelfth International Conference on Learning Representations, 2024

    work page 2024

  42. [42]

    Deep structural contrastive subspace clustering

    Bo Peng and Wenjie Zhu. Deep structural contrastive subspace clustering. InAsian Conference on Machine Learning, pp. 1145–1160. PMLR, 2021

    work page 2021

  43. [43]

    Deep residual matrix factorization for gait recognition

    Bo Peng, Wenjie Zhu, and Xiuhui Wang. Deep residual matrix factorization for gait recognition. In Proceedings of the 2020 12th International Conference on Machine Learning and Computing, pp. 330– 334, 2020

    work page 2020

  44. [44]

    Knowledge distillation with auxiliary variable

    Bo Peng, Zhen Fang, Guangquan Zhang, and Jie Lu. Knowledge distillation with auxiliary variable. In Forty-first International Conference on Machine Learning, 2024

    work page 2024

  45. [45]

    Conjnorm: Tractable density estima- tion for out-of-distribution detection

    Bo Peng, Yadan Luo, Yonggang Zhang, Yixuan Li, and Zhen Fang. Conjnorm: Tractable density estima- tion for out-of-distribution detection. InThe Twelfth International Conference on Learning Representa- tions, 2024

    work page 2024

  46. [46]

    An information-theoretical framework for under- standing out-of-distribution detection with pretrained vision-language models

    Bo Peng, Jie Lu, Guangquan Zhang, and Zhen Fang. An information-theoretical framework for under- standing out-of-distribution detection with pretrained vision-language models. InThe Thirty-ninth Annual Conference on Neural Information Processing Systems, 2025

    work page 2025

  47. [47]

    On the provable importance of gradients for au- tonomous language-assisted image clustering

    Bo Peng, Jie Lu, Guangquan Zhang, and Zhen Fang. On the provable importance of gradients for au- tonomous language-assisted image clustering. InProceedings of the IEEE/CVF International Conference on Computer Vision, pp. 19805–19815, 2025

    work page 2025

  48. [48]

    On the Provable Importance of Gradients for Language-Assisted Image Clustering

    Bo Peng, Jie Lu, Guangquan Zhang, and Zhen Fang. On the provable importance of gradients for language-assisted image clustering.arXiv preprint arXiv:2510.16335, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  49. [49]

    Distributional prototype learning for out-of-distribution detection

    Bo Peng, Jie Lu, Yonggang Zhang, Guangquan Zhang, and Zhen Fang. Distributional prototype learning for out-of-distribution detection. InProceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining V . 1, pp. 1104–1114, 2025

    work page 2025

  50. [50]

    Learning transferable visual models from natural language supervision

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. Learning transferable visual models from natural language supervision. InInternational conference on machine learning, pp. 8748–8763. PMLR, 2021. 13

    work page 2021

  51. [51]

    Do imagenet classifiers generalize to imagenet? InInternational conference on machine learning, pp

    Benjamin Recht, Rebecca Roelofs, Ludwig Schmidt, and Vaishaal Shankar. Do imagenet classifiers generalize to imagenet? InInternational conference on machine learning, pp. 5389–5400. PMLR, 2019

    work page 2019

  52. [52]

    Detecting out-of-distribution examples with gram matri- ces

    Chandramouli Shama Sastry and Sageev Oore. Detecting out-of-distribution examples with gram matri- ces. InInternational Conference on Machine Learning, pp. 8491–8501. PMLR, 2020

    work page 2020

  53. [53]

    Out-of-distribution detection with deep nearest neighbors

    Yiyou Sun, Yifei Ming, Xiaojin Zhu, and Yixuan Li. Out-of-distribution detection with deep nearest neighbors. InInternational Conference on Machine Learning, pp. 20827–20840. PMLR, 2022

    work page 2022

  54. [54]

    Non-parametric outlier synthesis.arXiv preprint arXiv:2303.02966, 2023

    Leitian Tao, Xuefeng Du, Xiaojin Zhu, and Yixuan Li. Non-parametric outlier synthesis.arXiv preprint arXiv:2303.02966, 2023

    work page arXiv 2023

  55. [55]

    The inaturalist species classification and detection dataset

    Grant Van Horn, Oisin Mac Aodha, Yang Song, Yin Cui, Chen Sun, Alex Shepard, Hartwig Adam, Pietro Perona, and Serge Belongie. The inaturalist species classification and detection dataset. InProceedings of the IEEE conference on computer vision and pattern recognition, pp. 8769–8778, 2018

    work page 2018

  56. [56]

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

    Haoqi Wang, Zhizhong Li, Litong Feng, and Wayne Zhang. Vim: Out-of-distribution with virtual-logit matching. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp. 4921–4930, 2022

    work page 2022

  57. [57]

    Can multi-label classification networks know what they don’t know?Advances in Neural Information Processing Systems, 34:29074–29087, 2021

    Haoran Wang, Weitang Liu, Alex Bocchieri, and Yixuan Li. Can multi-label classification networks know what they don’t know?Advances in Neural Information Processing Systems, 34:29074–29087, 2021

    work page 2021

  58. [58]

    Clipn for zero-shot ood detection: Teaching clip to say no

    Hualiang Wang, Yi Li, Huifeng Yao, and Xiaomeng Li. Clipn for zero-shot ood detection: Teaching clip to say no. InProceedings of the IEEE/CVF International Conference on Computer Vision, pp. 1802–1812, 2023

    work page 2023

  59. [59]

    Sun database: Large- scale scene recognition from abbey to zoo

    Jianxiong Xiao, James Hays, Krista A Ehinger, Aude Oliva, and Antonio Torralba. Sun database: Large- scale scene recognition from abbey to zoo. In2010 IEEE computer society conference on computer vision and pattern recognition, pp. 3485–3492. IEEE, 2010

    work page 2010

  60. [60]

    Generalized out-of-distribution detection: A survey.arXiv preprint arXiv:2110.11334, 2021

    Jingkang Yang, Kaiyang Zhou, Yixuan Li, and Ziwei Liu. Generalized out-of-distribution detection: A survey.arXiv preprint arXiv:2110.11334, 2021

    work page arXiv 2021

  61. [61]

    Adaneg: Adaptive negative proxy guided ood detection with vision-language models.Advances in Neural Information Processing Systems, 37:38744–38768, 2024

    Yabin Zhang and Lei Zhang. Adaneg: Adaptive negative proxy guided ood detection with vision-language models.Advances in Neural Information Processing Systems, 37:38744–38768, 2024

    work page 2024

  62. [62]

    Lapt: Label-driven automated prompt tuning for ood detection with vision-language models

    Yabin Zhang, Wenjie Zhu, Chenhang He, and Lei Zhang. Lapt: Label-driven automated prompt tuning for ood detection with vision-language models. InEuropean Conference on Computer Vision, pp. 271–288. Springer, 2025

    work page 2025

  63. [63]

    Learning to shape in-distribution feature space for out-of-distribution detection.Advances in Neural Information Processing Systems, 37: 49384–49402, 2024

    Yonggang Zhang, Jie Lu, Bo Peng, Zhen Fang, and Yiu-ming Cheung. Learning to shape in-distribution feature space for out-of-distribution detection.Advances in Neural Information Processing Systems, 37: 49384–49402, 2024

    work page 2024

  64. [64]

    Decoupling maxlogit for out-of-distribution detection

    Zihan Zhang and Xiang Xiang. Decoupling maxlogit for out-of-distribution detection. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 3388–3397, 2023

    work page 2023

  65. [65]

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

    work page 2017

  66. [66]

    Community detection in multiplex networks by deep structure-preserving non-negative matrix factorization.Applied Intelligence, 55(1):26, 2025

    Qinli Zhou, Wenjie Zhu, Hao Chen, and Bo Peng. Community detection in multiplex networks by deep structure-preserving non-negative matrix factorization.Applied Intelligence, 55(1):26, 2025

    work page 2025

  67. [67]

    Sparse and low-rank regularized deep subspace clustering.Knowledge-Based Systems, 204:106199, 2020

    Wenjie Zhu and Bo Peng. Sparse and low-rank regularized deep subspace clustering.Knowledge-Based Systems, 204:106199, 2020

    work page 2020

  68. [68]

    Manifold-based aggregation clustering for unsupervised vehicle re- identification.Knowledge-Based Systems, 235:107624, 2022

    Wenjie Zhu and Bo Peng. Manifold-based aggregation clustering for unsupervised vehicle re- identification.Knowledge-Based Systems, 235:107624, 2022

    work page 2022

  69. [69]

    Query set centered sparse projection learning for set based image classification.Applied Intelligence, 50(10):3400–3411, 2020

    Wenjie Zhu, Bo Peng, Han Wu, and Binhao Wang. Query set centered sparse projection learning for set based image classification.Applied Intelligence, 50(10):3400–3411, 2020

    work page 2020

  70. [70]

    Self-supervised embedding for subspace clustering

    Wenjie Zhu, Bo Peng, and Chunchun Chen. Self-supervised embedding for subspace clustering. In Proceedings of the 30th ACM International Conference on Information & Knowledge Management, pp. 3687–3691, 2021

    work page 2021

  71. [71]

    Unified robust network embedding framework for community detection via extreme adversarial attacks.Information Sciences, 643:119200, 2023

    Wenjie Zhu, Chunchun Chen, and Bo Peng. Unified robust network embedding framework for community detection via extreme adversarial attacks.Information Sciences, 643:119200, 2023. 14

    work page 2023

  72. [72]

    Deep discriminative dictionary pair learning for image classification.Applied Intelligence, 53(19):22017–22030, 2023

    Wenjie Zhu, Bo Peng, Chunchun Chen, and Hao Chen. Deep discriminative dictionary pair learning for image classification.Applied Intelligence, 53(19):22017–22030, 2023

    work page 2023

  73. [73]

    Dual knowledge distillation on multiview pseudo labels for unsupervised person re-identification.IEEE Transactions on Multimedia, 26:7359–7371, 2024

    Wenjie Zhu, Bo Peng, and Wei Qi Yan. Dual knowledge distillation on multiview pseudo labels for unsupervised person re-identification.IEEE Transactions on Multimedia, 26:7359–7371, 2024

    work page 2024

  74. [74]

    Deep inductive and scalable subspace clustering via nonlocal contrastive self-distillation.IEEE Transactions on Circuits and Systems for Video Technology, 2025

    Wenjie Zhu, Bo Peng, and Wei Qi Yan. Deep inductive and scalable subspace clustering via nonlocal contrastive self-distillation.IEEE Transactions on Circuits and Systems for Video Technology, 2025

    work page 2025

  75. [75]

    Mood 2020: A public benchmark for out-of- distribution detection and localization on medical images.IEEE Transactions on Medical Imaging, 41 (10):2728–2738, 2022

    David Zimmerer, Peter M Full, Fabian Isensee, Paul J ¨ager, Tim Adler, Jens Petersen, Gregor K ¨ohler, Tobias Ross, Annika Reinke, Antanas Kascenas, et al. Mood 2020: A public benchmark for out-of- distribution detection and localization on medical images.IEEE Transactions on Medical Imaging, 41 (10):2728–2738, 2022. A Derivation of Theorem 1 We plug in t...

    work page 2020