Pith. sign in

REVIEW 4 major objections 2 minor 1 cited by

DP-NCB: Privacy Preserving Fair Bandits

T0 review · 4 major / 2 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read DP-NCB is a single bandit framework that enforces $\epsilon$-differential privacy and attains order-optimal Nash regret under both global and local privacy models, without knowing the horizon in advance.

desk verdict The abstract claims a genuinely important simultaneous privacy-fairness result, but the supplied full text is an unrelated LLM watermarking paper, so none of the math or experiments can be checked from this package. read the letter →

arxiv 2508.03836 v1 pith:KDIB2B3J submitted 2025-08-05 cs.LG stat.ML

Dhruv Sarkar , Nishant Pandey , Sayak Ray Chowdhury This is my paper
classification cs.LGstat.ML
keywords differentialprivacymulti-armedbanditsNashregretfairnessgloballocalanytimealgorithmconfidencebound
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

DP-NCB is a proposed algorithm for the multi-armed bandit problem that aims to settle a previously open question: can a bandit learner protect user privacy and treat arms fairly at the same time? The paper claims that the answer is yes, and that both goals can be achieved without sacrificing the best possible learning rate. Concretely, DP-NCB is shown to satisfy $\epsilon$-differential privacy and to attain Nash regret matching known lower bounds up to logarithmic factors, in both the global and local privacy models, with no need to know the time horizon in advance. If true, the result gives a principled template for high-stakes sequential decisions, such as clinical trials or personalized services, where both privacy and equitable treatment matter.

What carries the argument

The central object is Nash regret itself, defined as the maximum over arms of the reward lost by that arm because it was not pulled often enough, combined with a confidence-bound exploration rule that is made differentially private. The mechanism carries the argument by ensuring that the private noise and the confidence radii are balanced so that no single arm is starved beyond the optimal order, while the $\epsilon$-privacy guarantee holds regardless of which arms the learner chooses. The anytime property follows from confidence bounds that do not require the horizon as an input.

What would settle it

Check whether the algorithm's confidence radius or privacy budget allocation ever uses the horizon $T$ as an input; if it does, the claim that DP-NCB is anytime is refuted by inspection, and the claimed regret bound would require knowing $T$ in advance.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that privacy and fairness are not competing constraints in sequential decision-making. The author introduces Differentially Private Nash Confidence Bound, a unified framework built on confidence-bound exploration whose private statistics are calibrated so that the worst-off arm's lost reward stays within the optimal order. The framework simultaneously delivers $\epsilon$-differential privacy and order-optimal Nash regret under both global and local differential privacy, and it is anytime, meaning the learner does not need to know the horizon in advance. Simulations on synthetic bandit instances are reported to show substantially lower Nash regret than existing baselines.

Load-bearing premise

The proof of optimality relies on the cited Nash-regret lower bounds being proven for exactly the same model, with the same reward distributions, the same differential privacy mechanism class, and the same fairness definition, so that matching them up to log factors really means optimal.

Editorial extensions

If this is right

  • A deployer can now request both $\epsilon$-differential privacy and a fairness guarantee from a bandit algorithm without paying more than logarithmic factors in regret.
  • The same algorithm works under global differential privacy, with a trusted curator, and local differential privacy, with privacy preserved at the user's device, so the result covers two common deployment architectures.
  • Because the algorithm is anytime, it can be stopped at arbitrary times or used with an unknown total number of rounds, which is typical in clinical and online settings.
  • Synthetic experiments suggest that the theoretical gains carry over to finite samples, with DP-NCB incurring lower Nash regret than state-of-the-art baselines.

Reading between the lines

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

  • A natural extension the paper does not test is whether the same design still matches lower bounds under other fairness objectives, such as max-min or proportional fairness, since the confidence-bound structure is tied to Nash regret.
  • The optimality claim inherits the strength of the cited lower bounds; if those bounds were proven under slightly different reward or privacy assumptions, the 'up to logarithmic factors' conclusion would need re-checking rather than carrying over automatically.
  • The anytime property could be stress-tested by changing the horizon partway through a run; the paper's simulations do not include this, but the anytime guarantee predicts only a polylogarithmic penalty.
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

4 major / 2 minor

Summary. The manuscript as provided consists only of the abstract of arXiv:2508.03836, which introduces DP-NCB, a differentially private Nash confidence bound algorithm for multi-armed bandits. The abstract claims that DP-NCB simultaneously guarantees epsilon-differential privacy and order-optimal Nash regret matching known lower bounds up to logarithmic factors, under both global and local differential privacy, without requiring prior knowledge of the time horizon, and that synthetic simulations show substantially lower Nash regret than state-of-the-art baselines. The body of the paper that would define the algorithm, state and prove the regret and privacy guarantees, specify the lower-bound comparison class, and describe the experiments is not included; the supplied full text is an unrelated manuscript on LLM watermarking.

Significance. If the abstract's claims were fully substantiated, the result would be a significant advance: it would resolve the open question of whether privacy and fairness can be achieved simultaneously in bandits, and it would offer a unified treatment of global and local differential privacy in fair bandits with an anytime guarantee. The claimed order-optimal Nash regret, matching lower bounds up to logarithmic factors, would be a strong and falsifiable theoretical statement, and the framework would be relevant to high-stakes applications. However, because no proof, algorithm description, lower-bound statement, or experimental protocol is available for scrutiny, the significance cannot currently be verified. I credit the authors for a clearly stated and plausible central claim, and I do not see internal inconsistency or circularity in the abstract alone; the problem is the absence of evidence, not an evident error.

major comments (4)
  1. [Full text] The supplied full text is not the manuscript under review; it is an unrelated paper on majority bit-aware LLM watermarking. No definition of DP-NCB, no pseudocode, no proof of the regret bound, no proof of the differential privacy guarantee, and no statement of the lower bounds appears anywhere in the available material. This makes the central claims of the paper unverifiable and is a load-bearing defect that no local revision can repair.
  2. [Abstract] The abstract's claim of matching known lower bounds up to logarithmic factors cannot be checked without the precise lower-bound statements. The comparison class must specify whether the lower bounds are for central (joint) differential privacy or local differential privacy, the per-round versus total privacy budget, the reward distribution class, and the exact definition of Nash regret. If the cited lower bounds are the non-private Nash-regret bounds from the fairness literature, the claim would be internally inconsistent, because local-DP bandits suffer unavoidable epsilon-dependent penalties; the manuscript provides no way to resolve this concern.
  3. [Abstract] The anytime claim requires a regret bound that holds uniformly over all horizons with no prior knowledge of T, together with a composition argument that accounts for an unbounded number of rounds. Neither the algorithm nor the composition analysis is present, so the anytime and differential privacy guarantees cannot be assessed.
  4. [Abstract] The empirical claim of substantially lower Nash regret than state-of-the-art baselines is summarized without any protocol: no instance generation procedure, no baseline list, no confidence intervals or error bars, no number of runs, and no code release. As a result, the simulations provide no independent support for the theoretical claims.
minor comments (2)
  1. [Full text] The title and abstract of the submitted package refer to a cs.LG paper on DP-NCB, while the appended full text is a cs.CL paper on watermarking; the packaging error should be corrected in any resubmission.
  2. [Abstract] The abstract should state whether the epsilon in epsilon-differential privacy is the same parameter under the global and local models and whether it is a per-round or total privacy budget, since these choices materially affect the meaning of the claimed optimality.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity is identifiable; the available text does not contain a fitted-input or self-referential derivation chain.

full rationale

The only target-paper content available is the abstract of arXiv:2508.03836 (DP-NCB). The abstract claims that DP-NCB simultaneously ensures epsilon-differential privacy and achieves order-optimal Nash regret, matching known lower bounds up to logarithmic factors. This claim is asserted against external lower bounds, not fitted to data, and no parameter in the abstract is defined in terms of the quantity being predicted. There is no self-citation, no uniqueness theorem invoked from the authors' own prior work, and no ansatz smuggled in via citation on the basis of the supplied text. The full text provided in the package is an unrelated paper on LLM watermarking, so the DP-NCB derivation chain, algorithm construction, and comparison class of lower bounds cannot be inspected from the available material. However, absence of the full text is not itself a circularity, and the instructions require exhibiting a specific reduction (e.g., Eq. X = Eq. Y by construction) before flagging circularity. No such reduction can be exhibited from the abstract alone. Therefore the honest finding is no significant circularity, with score 0.

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

Abstract-only review. No free parameters or invented entities are identifiable from the abstract. The stated order-optimality relies on prior lower bounds and on Nash regret as the fairness metric.

assumptions (2)
  • domain assumption The cited lower bounds for Nash regret in bandits are correct and apply to DP-NCB's exact problem setting.
    The abstract's 'matching known lower bounds up to logarithmic factors' assumes these bounds exist and are the right comparison class.
  • domain assumption Nash regret is the appropriate formalization of fairness in this bandit setting.
    The whole framing of fairness is inherited from prior fair-bandits work; the abstract does not justify this choice.

how reviews work

0 comments
Cite this review

Pith. "Pith review of DP-NCB: Privacy Preserving Fair Bandits." pith.science (2026). https://pith.science/paper/KDIB2B3J

@misc{pith2026250803836,
  author       = {Pith},
  title        = {Pith review of: DP-NCB: Privacy Preserving Fair Bandits},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/KDIB2B3J}},
  note         = {Machine review of arXiv:2508.03836}
}
abstract

Multi-armed bandit algorithms are fundamental tools for sequential decision-making under uncertainty, with widespread applications across domains such as clinical trials and personalized decision-making. As bandit algorithms are increasingly deployed in these socially sensitive settings, it becomes critical to protect user data privacy and ensure fair treatment across decision rounds. While prior work has independently addressed privacy and fairness in bandit settings, the question of whether both objectives can be achieved simultaneously has remained largely open. Existing privacy-preserving bandit algorithms typically optimize average regret, a utilitarian measure, whereas fairness-aware approaches focus on minimizing Nash regret, which penalizes inequitable reward distributions, but often disregard privacy concerns. To bridge this gap, we introduce Differentially Private Nash Confidence Bound (DP-NCB)-a novel and unified algorithmic framework that simultaneously ensures $\epsilon$-differential privacy and achieves order-optimal Nash regret, matching known lower bounds up to logarithmic factors. The framework is sufficiently general to operate under both global and local differential privacy models, and is anytime, requiring no prior knowledge of the time horizon. We support our theoretical guarantees with simulations on synthetic bandit instances, showing that DP-NCB incurs substantially lower Nash regret than state-of-the-art baselines. Our results offer a principled foundation for designing bandit algorithms that are both privacy-preserving and fair, making them suitable for high-stakes, socially impactful applications.

Discussion (0). Sign in to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Improved Algorithms for Nash Welfare in Linear Bandits

    cs.LG 2026-01 conditional novelty 7.0 of 10

    FairLinBandit achieves order-optimal Nash regret Õ(d/√T) and the first sublinear p-mean regret bounds in linear bandits for every real p.

Reference graph

Works this paper leans on

46 extracted references · 27 canonical work pages · cited by 1 Pith paper

  1. [1]

    Random oracles are practical: A paradigm for designing efficient protocols

    Mihir Bellare and Phillip Rogaway. Random oracles are practical: A paradigm for designing efficient protocols. InProceedings of the 1st ACM Conference on Computer and Communications Security, pages 62–73, 1993

    work page 1993

  2. [2]

    PostMark: A robust blackbox watermark for large language models

    Yapei Chang, Kalpesh Krishna, Amir Houmansadr, John Frederick Wieting, and Mohit Iyyer. PostMark: A robust blackbox watermark for large language models. In Yaser Al-Onaizan, Mohit Bansal, and Yun-Nung Chen, editors,Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, pages 8969–8987, Miami, Florida, USA, November 2024. A...

    work page 2024

  3. [3]

    Undetectable watermarks for language models

    Miranda Christ, Sam Gunn, and Or Zamir. Undetectable watermarks for language models. InThe Thirty Seventh Annual Conference on Learning Theory, pages 1125–1139. PMLR, 2024

    2024

  4. [4]

    Simmark: A robust sentence-level similarity-based water- marking algorithm for large language models.arXiv preprint arXiv:2502.02787, 2025

    Amirhossein Dabiriaghdam and Lele Wang. Simmark: A robust sentence-level similarity-based water- marking algorithm for large language models.arXiv preprint arXiv:2502.02787, 2025

    arXiv 2025

  5. [5]

    Hierarchical neural story generation

    Angela Fan, Mike Lewis, and Yann Dauphin. Hierarchical neural story generation. In Iryna Gurevych and Yusuke Miyao, editors,Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 889–898, Melbourne, Australia, July 2018. Association for Computational Linguistics

    2018

  6. [6]

    John Wiley & Sons, 1991

    William Feller.An introduction to probability theory and its applications, Volume 2, volume 2. John Wiley & Sons, 1991

    work page 1991

  7. [7]

    Bimark: Unbiased multilayer watermarking for large language models

    Xiaoyan Feng, He Zhang, Yanjun Zhang, Leo Yu Zhang, and Shirui Pan. Bimark: Unbiased multilayer watermarking for large language models. InForty-second International Conference on Machine Learning, 2025

    work page 2025

  8. [8]

    Three bricks to consolidate watermarks for large language models

    Pierre Fernandez, Antoine Chaffin, Karim Tit, Vivien Chappelier, and Teddy Furon. Three bricks to consolidate watermarks for large language models. In2023 IEEE International Workshop on Information Forensics and Security (WIFS), pages 1–6. IEEE, 2023

    work page 2023

Show all 46 references
  1. [9]

    WaterMax: breaking the LLM watermark detectability-robustness-quality trade-off.Advances in Neural Information Processing Systems, 37:18848–18881, 2024

    Eva Giboulot and Teddy Furon. WaterMax: breaking the LLM watermark detectability-robustness-quality trade-off.Advances in Neural Information Processing Systems, 37:18848–18881, 2024

    2024

  2. [10]

    The llama 3 herd of models

    Aaron Grattafiori, Abhimanyu Dubey, Abhinav Jauhri, Abhinav Pandey, Abhishek Kadian, Ahmad Al- Dahle, Aiesha Letman, Akhil Mathur, Alan Schelten, Alex Vaughan, et al. The llama 3 herd of models. arXiv preprint arXiv:2407.21783, 2024

    2024 arXiv

  3. [11]

    Teaching machines to read and comprehend.Advances in neural information processing systems, 28, 2015

    Karl Moritz Hermann, Tomas Kocisky, Edward Grefenstette, Lasse Espeholt, Will Kay, Mustafa Suleyman, and Phil Blunsom. Teaching machines to read and comprehend.Advances in neural information processing systems, 28, 2015

    2015

  4. [12]

    Semstamp: A semantic watermark with paraphrastic robustness for text generation

    Abe Hou, Jingyu Zhang, Tianxing He, Yichen Wang, Yung-Sung Chuang, Hongwei Wang, Lingfeng Shen, Benjamin Van Durme, Daniel Khashabi, and Yulia Tsvetkov. Semstamp: A semantic watermark with paraphrastic robustness for text generation. InProceedings of the 2024 Conference of the...

    2024

  5. [13]

    Stealthink: A multi-bit and stealthy watermark for large language models

    Ya Jiang, Chuxiong Wu, Massieh Kordi Boroujeny, Brian Mark, and Kai Zeng. Stealthink: A multi-bit and stealthy watermark for large language models. InForty-second International Conference on Machine Learning, 2025. 10

    2025

  6. [14]

    Watermark stealing in large language models

    Nikola Jovanovi´c, Robin Staab, and Martin Vechev. Watermark stealing in large language models. In Ruslan Salakhutdinov, Zico Kolter, Katherine Heller, Adrian Weller, Nuria Oliver, Jonathan Scarlett, and Felix Berkenkamp, editors,Proceedings of the 41st International Conferenc...

    2024

  7. [15]

    A watermark for large language models

    John Kirchenbauer, Jonas Geiping, Yuxin Wen, Jonathan Katz, Ian Miers, and Tom Goldstein. A watermark for large language models. InInternational Conference on Machine Learning, pages 17061–17084. PMLR, 2023

    2023

  8. [16]

    On the reliability of watermarks for large language models

    John Kirchenbauer, Jonas Geiping, Yuxin Wen, Manli Shu, Khalid Saifullah, Kezhi Kong, Kasun Fernando, Aniruddha Saha, Micah Goldblum, and Tom Goldstein. On the reliability of watermarks for large language models. InThe Twelfth International Conference on Learning Representations, 2024

    2024

  9. [17]

    Paraphrasing evades detectors of ai-generated text, but retrieval is an effective defense.Advances in Neural Information Processing Systems, 36:27469–27500, 2023

    Kalpesh Krishna, Yixiao Song, Marzena Karpinska, John Wieting, and Mohit Iyyer. Paraphrasing evades detectors of ai-generated text, but retrieval is an effective defense.Advances in Neural Information Processing Systems, 36:27469–27500, 2023

    2023

  10. [18]

    Robust distortion-free water- marks for language models.Transactions on Machine Learning Research, 2024

    Rohith Kuditipudi, John Thickstun, Tatsunori Hashimoto, and Percy Liang. Robust distortion-free water- marks for language models.Transactions on Machine Learning Research, 2024

    2024

  11. [19]

    Where am I from? identifying origin of LLM-generated content

    Liying Li, Yihan Bai, and Minhao Cheng. Where am I from? identifying origin of LLM-generated content. In Yaser Al-Onaizan, Mohit Bansal, and Yun-Nung Chen, editors,Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, pages 12218–12229, Miami,...

    2024

  12. [20]

    Aiwei Liu, Leyi Pan, Xuming Hu, Shuang Li, Lijie Wen, Irwin King, and Philip S. Yu. An unforgeable publicly verifiable watermark for large language models. InThe Twelfth International Conference on Learning Representations, 2024

    2024

  13. [21]

    A semantic invariant robust watermark for large language models

    Aiwei Liu, Leyi Pan, Xuming Hu, Shiao Meng, and Lijie Wen. A semantic invariant robust watermark for large language models. InThe Twelfth International Conference on Learning Representations, 2024

    2024

  14. [22]

    A survey of text watermarking in the era of large language models.ACM Computing Surveys, 57(2):1–36, 2024

    Aiwei Liu, Leyi Pan, Yijian Lu, Jingjing Li, Xuming Hu, Xi Zhang, Lijie Wen, Irwin King, Hui Xiong, and Philip Yu. A survey of text watermarking in the era of large language models.ACM Computing Surveys, 57(2):1–36, 2024

    2024

  15. [23]

    Distortion agnostic deep watermarking

    Xiyang Luo, Ruohan Zhan, Huiwen Chang, Feng Yang, and Peyman Milanfar. Distortion agnostic deep watermarking. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 13548–13557, 2020

    2020

  16. [24]

    Some methods of classification and analysis of multivariate observations

    James B McQueen. Some methods of classification and analysis of multivariate observations. InProc. of 5th Berkeley Symposium on Math. Stat. and Prob., pages 281–297, 1967

    1967

  17. [25]

    Modern hierarchical, agglomerative clustering algorithms.arXiv preprint arXiv:1109.2378, 2011

    Daniel Müllner. Modern hierarchical, agglomerative clustering algorithms.arXiv preprint arXiv:1109.2378, 2011

    2011 arXiv

  18. [26]

    Deeptextmark: a deep learning-driven text watermarking approach for identifying large language model generated text.IEEE Access, 12:40508–40520, 2024

    Travis Munyer, Abdullah All Tanvir, Arjon Das, and Xin Zhong. Deeptextmark: a deep learning-driven text watermarking approach for identifying large language model generated text.IEEE Access, 12:40508–40520, 2024

    2024

  19. [27]

    MarkLLM: An open-source toolkit for LLM watermarking

    Leyi Pan, Aiwei Liu, Zhiwei He, Zitian Gao, Xuandong Zhao, Yijian Lu, Binglin Zhou, Shuliang Liu, Xuming Hu, Lijie Wen, et al. MarkLLM: An open-source toolkit for LLM watermarking. InProceedings of the 2024 Conference on Empirical Methods in Natural Language Processing: System...

    2024

  20. [28]

    Pedregosa, G

    F. Pedregosa, G. Varoquaux, A. Gramfort, V . Michel, B. Thirion, O. Grisel, M. Blondel, P. Prettenhofer, R. Weiss, V . Dubourg, J. Vanderplas, A. Passos, D. Cournapeau, M. Brucher, M. Perrot, and E. Duchesnay. Scikit-learn: Machine learning in Python.Journal of Machine Learnin...

    2011

  21. [29]

    Markmywords: Analyzing and evaluating language model watermarks

    Julien Piet, Chawin Sitawarin, Vivian Fang, Norman Mu, and David Wagner. Markmywords: Analyzing and evaluating language model watermarks. In2025 IEEE Conference on Secure and Trustworthy Machine Learning (SaTML), pages 68–91. IEEE, 2025

    2025

  22. [30]

    Provably robust multi-bit watermarking for ai-generated text.arXiv preprint arXiv:2401.16820, 2024

    Wenjie Qu, Wengrui Zheng, Tianyang Tao, Dong Yin, Yanze Jiang, Zhihua Tian, Wei Zou, Jinyuan Jia, and Jiaheng Zhang. Provably robust multi-bit watermarking for ai-generated text.arXiv preprint arXiv:2401.16820, 2024. 11

    2024 arXiv

  23. [31]

    Exploring the limits of transfer learning with a unified text-to-text transformer

    Colin Raffel, Noam Shazeer, Adam Roberts, Katherine Lee, Sharan Narang, Michael Matena, Yanqi Zhou, Wei Li, and Peter J Liu. Exploring the limits of transfer learning with a unified text-to-text transformer. Journal of machine learning research, 21(140):1–67, 2020

    2020

  24. [32]

    Gaussian mixture models

    Douglas Reynolds. Gaussian mixture models. InEncyclopedia of biometrics, pages 827–832. Springer, 2015

    2015

  25. [33]

    Essays with instructions dataset

    Christoph Schuhmann. Essays with instructions dataset. https://huggingface.co/datasets/ ChristophSchuhmann/essays-with-instructions, 2023. Accessed: 2025-07-14

    2023

  26. [34]

    Gemma 2: Improving open language models at a practical size.arXiv preprint arXiv:2408.00118, 2024

    Gemma Team, Morgane Riviere, Shreya Pathak, Pier Giuseppe Sessa, Cassidy Hardin, Surya Bhupatiraju, Léonard Hussenot, Thomas Mesnard, Bobak Shahriari, Alexandre Ramé, et al. Gemma 2: Improving open language models at a practical size.arXiv preprint arXiv:2408.00118, 2024

    2024 arXiv

  27. [35]

    Llama 2: Open foundation and fine-tuned chat models.arXiv preprint arXiv:2307.09288, 2023

    Hugo Touvron, Louis Martin, Kevin Stone, Peter Albert, Amjad Almahairi, Yasmine Babaei, Nikolay Bashlykov, Soumya Batra, Prajjwal Bhargava, Shruti Bhosale, et al. Llama 2: Open foundation and fine-tuned chat models.arXiv preprint arXiv:2307.09288, 2023

    2023 arXiv

  28. [36]

    Towards codable watermarking for injecting multi-bits information to LLMs

    Lean Wang, Wenkai Yang, Deli Chen, Hao Zhou, Yankai Lin, Fandong Meng, Jie Zhou, and Xu Sun. Towards codable watermarking for injecting multi-bits information to LLMs. InThe Twelfth International Conference on Learning Representations, ICLR 2024, Vienna, Austria, May 7-11, 202...

    2024

  29. [37]

    Morphmark: Flexible adaptive watermarking for large language models.arXiv preprint arXiv:2505.11541, 2025

    Zongqi Wang, Tianle Gu, Baoyuan Wu, and Yujiu Yang. Morphmark: Flexible adaptive watermarking for large language models.arXiv preprint arXiv:2505.11541, 2025

    2025 arXiv

  30. [38]

    A survey on LLM-generated text detection: Necessity, methods, and future directions.Computational Linguistics, 51(1):275–338, 2025

    Junchao Wu, Shu Yang, Runzhe Zhan, Yulin Yuan, Lidia Sam Chao, and Derek Fai Wong. A survey on LLM-generated text detection: Necessity, methods, and future directions.Computational Linguistics, 51(1):275–338, 2025

    2025

  31. [39]

    Qwen3 technical report, 2025

    An Yang, Anfeng Li, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chang Gao, Chengen Huang, Chenxu Lv, Chujie Zheng, Dayiheng Liu, Fan Zhou, Fei Huang, Feng Hu, Hao Ge, Haoran Wei, Huan Lin, Jialong Tang, Jian Yang, Jianhong Tu, Jianwei Zhang, Jianxin Yang, Jia...

    2025

  32. [40]

    Qwen2 technical report.arXiv preprint arXiv:2407.10671, 2024

    An Yang, Baosong Yang, Binyuan Hui, Bo Zheng, Bowen Yu, Chang Zhou, Chengpeng Li, Chengyuan Li, Dayiheng Liu, Fei Huang, Guanting Dong, Haoran Wei, Huan Lin, Jialong Tang, Jialin Wang, Jian Yang, Jianhong Tu, Jianwei Zhang, Jianxin Ma, Jin Xu, Jingren Zhou, Jinze Bai, Jinzheng...

    2024 arXiv

  33. [41]

    Tracing text provenance via context-aware lexical substitution

    Xi Yang, Jie Zhang, Kejiang Chen, Weiming Zhang, Zehua Ma, Feng Wang, and Nenghai Yu. Tracing text provenance via context-aware lexical substitution. InProceedings of the AAAI Conference on Artificial Intelligence, volume 36, pages 11613–11621, 2022

    2022

  34. [42]

    Advancing beyond identification: Multi-bit watermark for large language models

    KiYoon Yoo, Wonhyuk Ahn, and Nojun Kwak. Advancing beyond identification: Multi-bit watermark for large language models. In Kevin Duh, Helena Gomez, and Steven Bethard, editors,Proceedings of the 2024 Conference of the North American Chapter of the Association for Computationa...

    2024

  35. [43]

    REMARK-LLM: A robust and efficient watermarking framework for generative large language models

    Ruisi Zhang, Shehzeen Samarah Hussain, Paarth Neekhara, and Farinaz Koushanfar. REMARK-LLM: A robust and efficient watermarking framework for generative large language models. In33rd USENIX Security Symposium (USENIX Security 24), pages 1813–1830, 2024

    2024

  36. [44]

    Character-level perturbations disrupt LLM watermarks.arXiv preprint arXiv:2509.09112, 2025

    Zhaoxi Zhang, Xiaomei Zhang, Yanjun Zhang, He Zhang, Shirui Pan, Bo Liu, Asif Qumer Gill, and Leo Yu Zhang. Character-level perturbations disrupt LLM watermarks.arXiv preprint arXiv:2509.09112, 2025

    2025 arXiv

  37. [45]

    The Mamba Mentality

    Jiren Zhu, Russell Kaplan, Justin Johnson, and Li Fei-Fei. Hidden: Hiding data with deep networks. In Proceedings of the European conference on computer vision (ECCV), pages 657–672, 2018. 12 Table 5: Notation table. Symbol Description γThe green list ratio mThe binary multi-b...

    2018

  38. [2024]

    Association for Computational Linguistics

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.