Student or educator? Sign up with your school email and get 6 months of Pro free.
CoChat
Sign Up
Cited by 36

The DOI matches its Crossref record; 15 of the 15 references it cites were found in a registry or an index.

Hardening LLM Fine-Tuning: From Differentially Private Data Selection to Trustworthy Model Quantization

Deng, Zehang, Sun, Ruoxi, Xue, Minhui, Ma, Wanlun, Wen, Sheng, Nepal, Surya, Xiang, Yang·2025·IEEE Transactions on Information Forensics and Security
DOI 10.1109/tifs.2025.3581103
Open this paper in CoChat — free account

Abstract

Critical infrastructures are increasingly integrating artificial intelligence (AI) technologies, including large language models (LLMs), into essential systems and services that are vital to societal functioning. Fine-tuning LLMs for specific domain tasks are crucial for their effective deployment in these contexts, but this process must carefully address both privacy and security concerns. Without proper safeguards, such integration can introduce additional risks, such as data leakage during training and diminished model trustworthiness due to the need for model compression to operate within limited bandwidth and computational capacity constraints. In this paper, we proposeHardening LLM Fine-tuning framework(HARDLLM), which addresses these challenges through two key components: (i) we develop a differentially private data selection method that ensures privacy protection by training the model exclusively on sampled and synthesized public data, thereby preventing any direct use of private data and enhancing leakage resilience throughout the training process, and (ii) we introduce a trustworthiness-aware model quantization approach to improve LLMs performance, such as reducing toxicity, enhancing adversarial robustness, and mitigating stereotypes, while maintaining negligible impact on model utility. Experimental results show that, the proposed algorithm ensures differential privacy when privacy budget is set at ϵ = 0.5, with only a 1% drop in accuracy, while other state-of-the…

Overview

Integrity signals

Registry

Crossref · 10.1109/tifs.2025.3581103

Cites

15 of 15 references verified0 citation errors · 0 not checkable

Statistics

Recomputing p-values needs the paper’s full text.

Verification

Claims the paper makes

This paper’s full text isn’t openly available, so its claims and self-check weren’t run. The registry and citation checks still apply.

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

Citations

Cites

Show all 15 references

Research works printed but not linked: 54

  1. 10.1109/MC.2020.3006177 · DOI
  2. 10.1109/MNET.2024.3384013 · DOI
  3. Hassanin, 2024, PLLM-CS: Pre-trained large language model (LLM) for cyber threat detection in satellite networks, arXiv:2405.05469 · arXiv
  4. 10.1109/JAS.2025.125498 · DOI
  5. 10.1109/JAS.2024.124983 · DOI
  6. 10.1109/JAS.2024.124971 · DOI
  7. 10.1109/TDSC.2024.3367737 · DOI
  8. Nayan, SoK: All you need to know about on-device ML model extraction—The gap between research and practice, 5233
  9. Mattioli, Identifying emerging cyber security threats and challenges for 2030, 64, 1
  10. 10.1109/tifs.2019.2932228 · DOI
  11. Hong, 2024, Decoding compressed trust: Scrutinizing the trustworthiness of efficient LLMs under compression, arXiv:2403.15447 · arXiv
  12. 10.1609/aaai.v36i8.20825 · DOI
  13. Xie, Data selection for language models via importance resampling, 34201
  14. Zhao, 2020, IDLG: Improved deep leakage from gradients, arXiv:2001.02610 · arXiv
  15. 10.18653/v1/2021.findings-emnlp.305 · DOI
  16. Dimitrov, LAMP: Extracting text from gradients with language model priors, 7641
  17. 10.18653/v1/2023.emnlp-main.765 · DOI
  18. 10.29012/jpc.880 · DOI
  19. Li, Large language models can be strong differentially private learners, 1
  20. 10.18653/v1/2023.acl-long.74 · DOI
  21. Xie, Differentially private synthetic data via foundation model APIS 2: Text, 54531
  22. Lin, AWQ: Activation-aware weight quantization for LLM compression and acceleration, 87
  23. 10.14722/spacesec.2024.23008 · DOI
  24. 10.1007/978-3-030-77772-2_14 · DOI
  25. 10.1038/s41598-021-01638-z · DOI
  26. 10.1080/15481603.2022.2143872 · DOI
  27. Maranto, 2024, LLMSat: A large language model-based goal-oriented agent for autonomous space exploration, arXiv:2405.01392 · arXiv
  28. Li, LoSparse: Structured compression of large language models based on low-rank and sparse approximation, 20336
  29. Gu, Minillm: Knowledge distillation of large language models, 1
  30. Dettmers, 2022, LLM.int8(): 8-bit matrix multiplication for transformers at scale, arXiv:2208.07339 · arXiv
  31. Shao, OmniQuant: Omnidirectionally calibrated quantization for large language models, 1
  32. Liu, QLLM: Accurate and efficient low-bitwidth quantization for large language models, 1
  33. Karumbunathan, 2022, Nvidia Jetson AGX Orin Series
  34. 10.1109/MM.2021.3061394 · DOI · Paper page
  35. 10.1145/2976749.2978318 · DOI · Paper page
  36. Lin, Rouge: A package for automatic evaluation of summaries, 74
  37. 10.1162/tacl_a_00290 · DOI · Paper page
  38. 10.1007/978-0-387-73003-5_196 · DOI
  39. Watson, On the importance of difficulty calibration in membership inference attacks, 1
  40. Douze, 2024, The faiss library, arXiv:2401.08281 · arXiv
  41. Gao, 2021, The pile: An 800GB dataset of diverse text for language modeling, arXiv:2101.00027 · arXiv
  42. Hendrycks, 2020, Measuring massive multitask language understanding, arXiv:2009.03300 · arXiv
  43. 10.18653/v1/D19-1410 · DOI · Paper page
  44. Xue, 2020, MT5: A massively multilingual pre-trained text-to-text transformer, arXiv:2010.11934 · arXiv
  45. 10.1038/s41746-024-01029-4 · DOI
  46. Bolya, Token merging: Your ViT but faster, 1
  47. 10.1561/0400000042 · DOI · Paper page
  48. 10.18653/v1/D13-1170 · DOI · Paper page
  49. Hou, 2024, Bridging language and items for retrieval and recommendation, arXiv:2403.03952 · arXiv
  50. Zhang, Character-level convolutional networks for text classification, 1
  51. 10.18653/v1/2021.findings-acl.449 · DOI
  52. 10.18653/v1/2020.coling-main.581 · DOI
  53. Touvron, 2023, Llama2: Open foundation and fine-tuned chat models, arXiv:2307.09288 · arXiv
  54. 10.1016/j.procs.2020.11.007 · DOI

Cited by

See all 36

Metadata

Authors
Deng, Z., Sun, R., Xue, M., Ma, W., Wen, S., Nepal, S., & Xiang, Y.
Venue
IEEE Transactions on Information Forensics and Security
Publisher
Institute of Electrical and Electronics Engineers (IEEE)
Published
2025
Volume
20
Pages
7211-7226
ISSN
  • 1556-6013
  • 1556-6021
DOI
10.1109/tifs.2025.3581103
Cited by
36
References
15
Licence

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: Hardening LLM Fine-Tuning: From Differentially Private Data Selection to Trustworthy Model Quantization. https://app.cochat.ai/papers/10.1109/tifs.2025.3581103
The paper
Deng, Z., Sun, R., Xue, M., Ma, W., Wen, S., Nepal, S., & Xiang, Y. (2025). Hardening LLM Fine-Tuning: From Differentially Private Data Selection to Trustworthy Model Quantization. IEEE Transactions on Information Forensics and Security. https://doi.org/10.1109/tifs.2025.3581103

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.