Pith. sign in

REVIEW 2 major objections 4 minor 49 references

Dataset Ownership in the Era of Large Language Models

T0 review · 2 major / 4 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read This survey claims the scattered field of dataset copyright protection can be organized by how much a method alters the data: non-intrusive detection, minimally-intrusive watermarking, and maximally-intrusive reversible adversarial examples

desk verdict A useful survey with a reasonable taxonomy that is undermined by a fixable internal inconsistency in its clean-label category and a missing citation; worth engaging after revision. read the letter →

arxiv 2509.05921 v1 pith:JE2UMPGI submitted 2025-09-07 cs.CR

Kun Li , Cheng Wang , Minghui Xu , Yue Zhang , Xiuzhen Cheng This is my paper · ORCID
classification cs.CR
keywords datasetownershipcopyrightprotectionwatermarkinglargelanguagemodelsreversibleadversarialexampleszero-watermarkingauditingmembershipinference
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

The paper is a survey claiming that all technical approaches to protecting dataset ownership fall into three categories based on how much they modify the data: non-intrusive methods that verify ownership without touching the data, minimally-intrusive methods that embed lightweight reversible signals, and maximally-intrusive methods that aggressively alter data to block unauthorized use. If this taxonomy holds, it gives researchers and practitioners a clear map of the field, showing that each approach balances data integrity against enforcement strength differently. The survey also argues that most current work targets image datasets, leaving text, audio, and other modalities underexplored, and that scalable, cross-domain, privacy-aware protection remains an open problem.

What carries the argument

The tripartite taxonomy itself is the load-bearing structure: it uses 'degree of data modification' as the organizing axis. Non-intrusive means no modification, minimally-intrusive means lightweight reversible changes, and maximally-intrusive means aggressive, usually reversible adversarial alterations. This axis lets the survey sort representative methods into bins and analyze each bin's trade-offs.

What would settle it

A side-by-side reimplementation of one method from each category on the same benchmark dataset would test whether the stated trade-offs actually separate by category; if a 'minimally-intrusive' method matches a 'maximally-intrusive' method on both robustness and model degradation, the boundary is arbitrary. More directly, finding any published protection method whose data modification is neither lightweight-reversible nor aggressive-adversarial would break the three-bin claim.

Watch

Extended reading notes

Core claim

The central claim is that dataset copyright protection methods can be systematically classified by their degree of data modification. Non-intrusive methods, such as zero-watermarking and decision-boundary or membership-inference detection, prove ownership without altering data. Minimally-intrusive methods, including reversible data hiding and dataset watermarking (backdoor, domain, and radioactive data), embed subtle signals that preserve most data quality. Maximally-intrusive methods, built on reversible adversarial examples, apply substantial perturbations so that unauthorized models fail while authorized users can recover the original data. The survey organizes the field along this single

Load-bearing premise

The survey's category boundaries are only as sound as its summaries of the methods it classifies, and one boundary is internally inconsistent: it defines clean-label watermarking as not altering labels, then includes backdoor methods that inject mislabeled data.

Editorial extensions

If this is right

  • A practitioner can classify any new protection method by asking how much it modifies the dataset, making the fragmented literature navigable by a single question.
  • The taxonomy implies that methods from different categories cannot be fairly compared on one metric: each bin optimizes a different balance of data preservation, robustness, and enforcement.
  • Since the survey finds most work targets image data, the same three categories become a template for exploring text, audio, and code datasets, where the field is comparatively empty.
  • The distinct category of maximally-intrusive methods shows that active enforcement through reversible adversarial examples is a separate paradigm from passive verification, not merely a stronger watermark.
  • The survey's own internal inconsistency in classifying backdoor watermarking as clean-label suggests that as methods combine mechanisms, category boundaries can blur and may need refinement.

Reading between the lines

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

  • If the taxonomy is correct, a practical decision procedure follows: choose a category by how much alteration the data owner tolerates and by whether the threat model is detection or active deterrence; the paper leaves this decision guide implicit.
  • The survey's observation that most methods target images hints that 'intrusiveness' may mean different things across modalities, e.g., reversible text watermarking may not map cleanly onto reversible pixel perturbation.
  • The clean-label/backdoor contradiction suggests the field is converging on methods that are minimally intrusive in data terms but maximally intrusive in behavioral terms, so a single modification axis may eventually need a second axis measuring behavioral enforcement.
  • A testable extension would be to apply one representative method from each category to the same benchmark dataset and check whether the predicted trade-offs (robustness vs fidelity vs scalability) actually separate by category; a mismatch would challenge the taxonomy's practical value.
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.

Referee Report

2 major / 4 minor

Summary. This survey reviews technical methods for dataset copyright protection in machine learning and organizes them into three categories based on the degree of data modification: non-intrusive methods (zero-watermarking, model-based detection), minimally-intrusive methods (reversible data hiding, clean-label/backdoor/domain/radioactive watermarking), and maximally-intrusive methods (reversible adversarial examples, visible watermark perturbations, Glaze). It also discusses open challenges including limited domain coverage, scalability, privacy, and potential future directions such as cross-domain protection and blockchain-based provenance.

Significance. The proposed three-category taxonomy is a plausible and useful organizing principle for a fragmented literature, and the survey covers a broad range of representative methods with a compact summary table. If the internal inconsistencies are corrected, the survey could serve as a valuable entry point for researchers and practitioners. The paper also correctly identifies that most existing work targets image datasets and that scalability, privacy, and LLM-specific protection remain underexplored. As a survey, it contains no machine-checked proofs or reproducibility artifacts; its contribution is organizational, and self-citations are present but do not determine the taxonomy.

major comments (2)
  1. [§5.2 and Table 1] The central taxonomy is undermined by an internal contradiction in the 'Clean-label watermarking' category. §5.2 defines clean-label watermarking as not directly altering labels, yet immediately classifies backdoor watermarking (Li et al. [20–22], Tang et al. [36]) in that category and then admits that backdoor methods 'often require injecting mislabeled data' and 'inherently introduce erroneous classification behaviors.' Table 1 lists backdoor watermarking under Clean-label watermarking. Because the categories are defined by degree of data modification, this inconsistency means a reader cannot reliably classify backdoor watermarking. Please restructure the category (e.g., separate poison-label and clean-label backdoor methods, or place backdoor watermarking in its own subcategory) and correct Table 1 accordingly.
  2. [Abstract, §3, §5.2, §6] The defining criterion for minimally-intrusive methods is stated as 'lightweight, reversible changes' (abstract) and 'often reversible signals' (§3). However, §5.2 describes dataset watermarking techniques—backdoor, domain, and radioactive watermarking—that permanently modify training samples or labels and provide no mechanism for recovering original data. RDH is reversible, but dataset watermarking is not. At the other extreme, §6 places Glaze, which 'add[s] minor perturbations to artworks,' in maximally-intrusive despite the category being defined by 'aggressive' alterations. The 'degree and nature of data modification' criterion is therefore applied inconsistently. Please define each category with explicit dimensions (e.g., permanence, label alteration, perturbation magnitude) and reclassify methods accordingly, or soften the reversibility and aggressiveness claims.
minor comments (4)
  1. [§4 Summary and Supplementary Note] The text cites 'Gupta et al.'s TrailChain [?]' with a missing reference placeholder. Please supply the full citation, or remove the claim if the work cannot be properly referenced.
  2. [References 39 and 40] References [39] and [40] are duplicates: both describe the same paper on reversible data hiding for color images based on prediction-error value ordering and adaptive embedding, with the same DOI. Please merge.
  3. [§1 Introduction] The roadmap in §1 lists Sections 2, 4, 5, 6, 7, and 8 but omits Section 3 (Overview), which appears in the paper. Please update the roadmap.
  4. [Title and §7] The title emphasizes the 'Era of Large Language Models,' but most described methods target image datasets; LLM-specific protection is mentioned only briefly. Consider either adding a dedicated discussion of LLM-specific dataset watermarking or tempering the title's scope.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity found: the survey is a descriptive taxonomy, not a derivation chain.

full rationale

This paper is a survey with no mathematical derivation or predictive chain: the tripartite taxonomy (non-/minimally-/maximally-intrusive) is a definitional organization by degree and nature of data modification, as stated in Section 3 ('categorizing them into three main paradigms based on the degree and nature of data modification involved'). No fitted parameter is later relabeled as a prediction, no uniqueness theorem is invoked to forbid alternatives, and no central claim reduces by construction to its own inputs. The self-citations ([44] in Section 1, [5] in Table 1/Section 6, [23] in Section 4) are contextual and non-load-bearing: they support background statements about privacy research, reversible adversarial examples, and blockchain provenance respectively, and do not determine the taxonomy. The internal tension in Section 5.2 between the clean-label definition ('does not directly alter the labels or minimizes such alterations') and the admission that backdoor watermarking methods 'often require injecting mislabeled data' is a real coherence problem for the proposed category boundaries, as is the missing citation 'Gupta et al.'s TrailChain [?]' in Section 4, but these are correctness/completeness issues, not circular reductions. Because the survey does not derive new results from its inputs, there is no equation-level or self-definitional circularity; the honest finding is score 0.

Assumptions & free parameters 0 free parameters · 2 assumptions · 0 invented entities

As a survey, the paper introduces no free parameters or invented entities. Its framework rests on domain assumptions about the usefulness of the modification-degree axis and about the representativeness of the selected references.

assumptions (2)
  • domain assumption The degree of data modification is a meaningful organizing axis for dataset protection methods.
    Section 3 introduces the tripartite division based on the degree and nature of data modification; no justification is given for why this axis is superior to alternatives such as attack model or data modality.
  • domain assumption The surveyed representative works are representative of the broader literature.
    Section 3 states the survey focuses on methods that exemplify distinct approaches rather than being exhaustive, but it provides no systematic inclusion criteria, making representativeness an unverified assumption.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Dataset Ownership in the Era of Large Language Models." pith.science (2026). https://pith.science/paper/JE2UMPGI

@misc{pith2026250905921,
  author       = {Pith},
  title        = {Pith review of: Dataset Ownership in the Era of Large Language Models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/JE2UMPGI}},
  note         = {Machine review of arXiv:2509.05921}
}
read the original abstract

As datasets become critical assets in modern machine learning systems, ensuring robust copyright protection has emerged as an urgent challenge. Traditional legal mechanisms often fail to address the technical complexities of digital data replication and unauthorized use, particularly in opaque or decentralized environments. This survey provides a comprehensive review of technical approaches for dataset copyright protection, systematically categorizing them into three main classes: non-intrusive methods, which detect unauthorized use without modifying data; minimally-intrusive methods, which embed lightweight, reversible changes to enable ownership verification; and maximally-intrusive methods, which apply aggressive data alterations, such as reversible adversarial examples, to enforce usage restrictions. We synthesize key techniques, analyze their strengths and limitations, and highlight open research challenges. This work offers an organized perspective on the current landscape and suggests future directions for developing unified, scalable, and ethically sound solutions to protect datasets in increasingly complex machine learning ecosystems.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

49 extracted references · 41 canonical work pages

  1. [36]

    ACM SIGKDD Explorations Newsletter25(1), 43–53 (2023)

    Tang, R., Feng, Q., Liu, N., Yang, F., Hu, X.: Did you train on my dataset? towards public dataset protection with cleanlabel backdoor watermarking. ACM SIGKDD Explorations Newsletter25(1), 43–53 (2023)

    work page 2023

  2. [1]

    High-Confidence Computing4(2), 100211 (2024)

    A survey on large language model (llm) security and privacy: The good, the bad, and the ugly. High-Confidence Computing4(2), 100211 (2024). https://doi.org/https://doi.org/10.1016/j.hcc.2024.100211, https://www.sciencedirect.com/science/article/pii/S266729522400014X

    arXiv 2024

  3. [2]

    Information12(11), 465 (2021)

    Asswad, J., Marx Gómez, J.: Data ownership: a survey. Information12(11), 465 (2021)

    work page 2021

  4. [3]

    In: Proceedings of the 2022 ACM on International Workshop on Security and Privacy Analytics

    Atli Tekgul, B.G., Asokan, N.: On the effectiveness of dataset watermarking. In: Proceedings of the 2022 ACM on International Workshop on Security and Privacy Analytics. pp. 93–99 (2022)

    work page 2022

  5. [4]

    IEEE transactions on cybernetics46(5), 1132–1143 (2015)

    Cao, X., Du, L., Wei, X., Meng, D., Guo, X.: High capacity reversible data hiding in encrypted images by patch-level sparse representation. IEEE transactions on cybernetics46(5), 1132–1143 (2015)

    work page 2015

  6. [5]

    In: 2024 International Joint Conference on Neural Networks (IJCNN)

    Cao, X., Liu, J., Yin, J., Cheng, X., Li, J., Ma, H., Luo, G.: Reversible adversarial examples based on self-embedding watermark for image privacy protection. In: 2024 International Joint Conference on Neural Networks (IJCNN). pp. 1–8. IEEE (2024)

    work page 2024

  7. [6]

    In: 2017 ieee symposium on security and privacy (sp)

    Carlini, N., Wagner, D.: Towards evaluating the robustness of neural networks. In: 2017 ieee symposium on security and privacy (sp). pp. 39–57. Ieee (2017)

    2017

  8. [7]

    arXiv preprint arXiv:2109.10870 (2021)

    Chandrasekaran, V., Jia, H., Thudi, A., Travers, A., Yaghini, M., Papernot, N.: Sok: Machine learning governance. arXiv preprint arXiv:2109.10870 (2021)

    arXiv 2021

Show all 49 references
  1. [8]

    Multimedia Tools and Applications83(4), 11215–11227 (2024)

    Chen, L., Zhu, S., Andrew, A., Yin, Z.: Reversible attack based on local visible adversarial perturbation. Multimedia Tools and Applications83(4), 11215–11227 (2024)

    2024

  2. [9]

    IEEE Internet of Things Journal (2024) Dataset Ownership in the Era of Large Language Models 13

    Chen, Z., Chai, X., Gan, Z., Wang, B., Zhang, Y.: Rae-vwp: a reversible adversarial example-based privacy and copyright protection method of medical images for internet of medical things. IEEE Internet of Things Journal (2024) Dataset Ownership in the Era of Large Language Models 13

    2024

  3. [10]

    arXiv preprint arXiv:2410.16618 (2024)

    Du, L., Zhou, X., Chen, M., Zhang, C., Su, Z., Cheng, P., Chen, J., Zhang, Z.: Sok: Dataset copyright auditing in machine learning systems. arXiv preprint arXiv:2410.16618 (2024)

    2024 arXiv

  4. [11]

    Dziedzic, A., Duan, H., Kaleem, M.A., Dhawan, N., Guan, J., Cattan, Y., Boenisch, F., Papernot, N.: Dataset inference for self-supervised models (2023), https://arxiv.org/abs/2209.09024

    2023 arXiv

  5. [12]

    In: International Conference of Pioneering Computer Scientists, Engineers and Educators

    Fang, Y., Jia, J., Yang, Y., Lyu, W.: Improving transferability reversible adver- sarial examples based on flipping transformation. In: International Conference of Pioneering Computer Scientists, Engineers and Educators. pp. 417–432. Springer (2023)

    2023

  6. [13]

    IEEE Transactions on Multimedia24, 2329–2338 (2021)

    Gao, L., Huang, Z., Song, J., Yang, Y., Shen, H.T.: Push & pull: Transferable adversarial examples with attentive attack. IEEE Transactions on Multimedia24, 2329–2338 (2021)

    2021

  7. [14]

    Advances in Neural Information Processing Systems36, 54421–54450 (2023)

    Guo, J., Li, Y., Wang, L., Xia, S.T., Huang, H., Liu, C., Li, B.: Domain watermark: Effective and harmless dataset copyright protection is closed at hand. Advances in Neural Information Processing Systems36, 54421–54450 (2023)

    2023

  8. [15]

    Hisamoto, S., Post, M., Duh, K.: Membership inference attacks on sequence-to- sequence models: Is my data in your machine translation system? Transactions of the Association for Computational Linguistics8, 49–63 (2020)

    2020

  9. [16]

    IEEE Transactions on Pattern Analysis and Machine Intelligence44(12), 10196–10208 (2022)

    Hu, R., Xiang, S.: Reversible data hiding by using cnn prediction and adaptive em- bedding. IEEE Transactions on Pattern Analysis and Machine Intelligence44(12), 10196–10208 (2022). https://doi.org/10.1109/TPAMI.2021.3131250

    2022

  10. [17]

    Image and Vi- sion Computing147, 105094 (2024)

    Huang, J., Huang, G., Zhang, X., Yuan, X., Xie, F., Pun, C.M., Zhong, G.: Black- box reversible adversarial examples with invertible neural network. Image and Vi- sion Computing147, 105094 (2024)

    2024

  11. [18]

    IEEE Transactions on Circuits and Systems for Video Technology30(8), 2353– 2365 (2020)

    Ke, Y., Zhang, M.Q., Liu, J., Su, T.T., Yang, X.Y.: Fully homomorphic encryption encapsulated difference expansion for reversible data hiding in encrypted domain. IEEE Transactions on Circuits and Systems for Video Technology30(8), 2353– 2365 (2020)

    2020

  12. [19]

    IEEE Transactions on In- formation forensics and security8(7), 1091–1100 (2013)

    Li, X., Zhang, W., Gui, X., Yang, B.: A novel reversible data hiding scheme based on two-dimensional difference-histogram modification. IEEE Transactions on In- formation forensics and security8(7), 1091–1100 (2013)

    2013

  13. [20]

    In: Oh, A.H., Agarwal, A., Belgrave, D., Cho, K

    Li, Y., Bai, Y., Jiang, Y., Yang, Y., Xia, S.T., Li, B.: Untargeted backdoor water- mark: Towards harmless and stealthy dataset copyright protection. In: Oh, A.H., Agarwal, A., Belgrave, D., Cho, K. (eds.) Advances in Neural Information Process- ing Systems (2022), https://ope...

    2022

  14. [21]

    Li, Y., Zhang, Z., Bai, J., Wu, B., Jiang, Y., Xia, S.T.: Open-sourced dataset protection via backdoor watermarking (2020), https://arxiv.org/abs/2010.05821

    2020 arXiv

  15. [22]

    IEEE Transactions on Information Forensics and Security18, 2318–2332 (2023)

    Li, Y., Zhu, M., Yang, X., Jiang, Y., Wei, T., Xia, S.T.: Black-box dataset own- ership verification via backdoor watermarking. IEEE Transactions on Information Forensics and Security18, 2318–2332 (2023)

    2023

  16. [23]

    IEEE Transactions on Computers71(12), 3268–3280 (2022)

    Liu, C., Guo, H., Xu, M., Wang, S., Yu, D., Yu, J., Cheng, X.: Extending on-chain trust to off-chain – trustworthy blockchain data collection using trusted execution environment (tee). IEEE Transactions on Computers71(12), 3268–3280 (2022). https://doi.org/10.1109/TC.2022.3148379

    2022

  17. [24]

    IEEE Transactions on Information Forensics and Security17, 1024–1037 (2022)

    Liu, G., Xu, T., Ma, X., Wang, C.: Your model trains on my data? protecting intellectual property of training data via membership fingerprint authentication. IEEE Transactions on Information Forensics and Security17, 1024–1037 (2022). https://doi.org/10.1109/TIFS.2022.3155921 ...

    2022

  18. [25]

    Pattern Recognition134, 109048 (2023)

    Liu, J., Zhang, W., Fukuchi, K., Akimoto, Y., Sakuma, J.: Unauthorized ai cannot recognize me: Reversible adversarial example. Pattern Recognition134, 109048 (2023)

    2023

  19. [26]

    Journal of Visual Communication and Image Rep- resentation97, 103935 (2023)

    Luo,J.,He,P.,Liu,J.,Wang,H.,Wu,C.,Zhou,S.:Reversibleadversarialsteganog- raphy for security enhancement. Journal of Visual Communication and Image Rep- resentation97, 103935 (2023)

    2023

  20. [27]

    Maini, P., Yaghini, M., Papernot, N.: Dataset inference: Ownership resolution in machine learning (2021), https://arxiv.org/abs/2104.10706

    2021 arXiv

  21. [28]

    In: 2018 IEEE European symposium on security and privacy (EuroS&P)

    Papernot, N., McDaniel, P., Sinha, A., Wellman, M.P.: Sok: Security and privacy in machine learning. In: 2018 IEEE European symposium on security and privacy (EuroS&P). pp. 399–414. IEEE (2018)

    2018

  22. [29]

    International Journal of Multimedia Information Retrieval 9(4), 249–270 (2020)

    Ray, A., Roy, S.: Recent trends in image watermarking techniques for copyright protection: a survey. International Journal of Multimedia Information Retrieval 9(4), 249–270 (2020)

    2020

  23. [30]

    Expert Systems with Applications226, 120217 (2023)

    Ren, N., Guo, S., Zhu, C., Hu, Y.: A zero-watermarking scheme based on spatial topological relations for vector dataset. Expert Systems with Applications226, 120217 (2023)

    2023

  24. [31]

    In: International Conference on Machine Learning

    Sablayrolles, A., Douze, M., Schmid, C., Jégou, H.: Radioactive data: tracing through training. In: International Conference on Machine Learning. pp. 8326–

  25. [32]

    In: 32nd USENIX Security Symposium (USENIX Security 23)

    Shan, S., Cryan, J., Wenger, E., Zheng, H., Hanocka, R., Zhao, B.Y.: Glaze: Pro- tecting artists from style mimicry by{Text-to-Image}models. In: 32nd USENIX Security Symposium (USENIX Security 23). pp. 2187–2204 (2023)

    2023

  26. [33]

    In: Proceedings of the ACM Turing Award Celebration Conference-China 2024

    Shang, Y., Xue, M., Zhang, L.Y., Zhang, Y., Liu, W.: Tracking the leaker: An encodable watermarking method for dataset intellectual property protection. In: Proceedings of the ACM Turing Award Celebration Conference-China 2024. pp. 114–119 (2024)

    2024

  27. [34]

    In: Proceedings of the 31st ACM Joint European Software Engineering Conference and Symposium on the Founda- tions of Software Engineering

    Sun, Z., Du, X., Song, F., Li, L.: Codemark: Imperceptible watermarking for code datasets against neural code completion models. In: Proceedings of the 31st ACM Joint European Software Engineering Conference and Symposium on the Founda- tions of Software Engineering. pp. 1561–...

    2023

  28. [35]

    Szegedy, C., Zaremba, W., Sutskever, I., Bruna, J., Erhan, D., Good- fellow, I., Fergus, R.: Intriguing properties of neural networks (2014), https://arxiv.org/abs/1312.6199

    2014 arXiv

  29. [37]

    IEEE Transactions on Information Forensics and Security18, 4116–4127 (2023)

    Tian, Z., Wang, Z., Abdelmoniem, A.M., Liu, G., Wang, C.: Knowledge represen- tation of training data with adversarial examples supporting decision boundary. IEEE Transactions on Information Forensics and Security18, 4116–4127 (2023). https://doi.org/10.1109/TIFS.2023.3293418

    2023

  30. [38]

    Information Sciences470, 109–120 (2019)

    Wang, C., Wang, X., Xia, Z., Zhang, C.: Ternary radial harmonic fourier moments based robust stereo image zero-watermarking algorithm. Information Sciences470, 109–120 (2019)

    2019

  31. [39]

    Journal of Visual Communication and Image Representation103, 104239 (2024)

    Wang, H., Wang, D., Chu, Z., Rao, Z., Yao, Y.: Reversible data hiding for color images based on prediction-error value ordering and adaptive embedding. Journal of Visual Communication and Image Representation103, 104239 (2024)

    2024

  32. [40]

    Wang, H., Wang, D., Chu, Z., Rao, Z., Yao, Y.: Reversible data hiding for color images based on prediction-error value or- dering and adaptive embedding. J. Vis. Comun. Image Repre- Dataset Ownership in the Era of Large Language Models 15 sent.103(C) (Aug 2024). https://doi.or...

    2024

  33. [41]

    IEEE Transactions on Information Forensics and Security (2024)

    Wei,C.,Wang,Y.,Gao,K.,Shao,S.,Li,Y.,Wang,Z.,Qin,Z.:Pointncbw:Towards dataset ownership verification for point clouds via negative clean-label backdoor watermark. IEEE Transactions on Information Forensics and Security (2024)

    2024

  34. [42]

    Pattern Recognition140, 109549 (2023)

    Xiong, L., Wu, Y., Yu, P., Zheng, Y.: A black-box reversible adversarial example for authorizable recognition to shared images. Pattern Recognition140, 109549 (2023)

    2023

  35. [43]

    Applied Intelligence53(6), 7298–7309 (2023)

    Xue, M., Wu, Y., Zhang, Y., Wang, J., Liu, W.: Dataset authorization control: protect the intellectual property of dataset via reversible feature space adversarial examples. Applied Intelligence53(6), 7298–7309 (2023)

    2023

  36. [44]

    High-Confidence Computing5(2), 100300 (2025)

    Yan, B., Li, K., Xu, M., Dong, Y., Zhang, Y., Ren, Z., Cheng, X.: On protecting the data privacy of large language models (llms) and llm agents: A literature review. High-Confidence Computing5(2), 100300 (2025). https://doi.org/https://doi.org/10.1016/j.hcc.2025.100300, https:...

    2025

  37. [45]

    High-Confidence Computing4(3), 100199 (2024)

    Yang, Z., Chen, X., He, Y., Liu, L., Che, Y., Wang, X., Xiao, K., Xu, G.: An attribute-based access control scheme using blockchain technology for iot data protection. High-Confidence Computing4(3), 100199 (2024). https://doi.org/https://doi.org/10.1016/j.hcc.2024.100199, http...

    2024

  38. [46]

    Pattern Recognition Letters 166, 1–7 (2023)

    Yin, Z., Chen, L., Lyu, W., Luo, B.: Reversible attack based on adversarial pertur- bation and reversible data hiding in yuv colorspace. Pattern Recognition Letters 166, 1–7 (2023)

    2023

  39. [47]

    arXiv preprint arXiv:1911.02360 (2019)

    Yin, Z., Wang, H., Chen, L., Wang, J., Zhang, W.: Reversible adversarial attack based on reversible image transformation. arXiv preprint arXiv:1911.02360 (2019)

    1911 arXiv

  40. [48]

    IEEE transactions on image processing22(7), 2775–2785 (2013)

    Zhang, W., Hu, X., Li, X., Yu, N.: Recursive histogram modification: establishing equivalency between reversible data hiding and lossless data compression. IEEE transactions on image processing22(7), 2775–2785 (2013)

    2013

  41. [49]

    IEEE Transactions on Information Forensics and Security17, 2468–2482 (2022)

    Zhao, M., Wang, B., Wang, W., Kong, Y., Zheng, T., Ren, K.: Guided erasable adversarial attack (geaa) toward shared data protection. IEEE Transactions on Information Forensics and Security17, 2468–2482 (2022)

    2022

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.