Conference paper · 2022 · Open access

Just Fine-tune Twice: Selective Differential Privacy for Large Language Models

Weiyan Shi, Ryan Shea, Si Chen, Chiyuan Zhang, Ruoxi Jia, Yu Zhou

Conference on Empirical Methods in Natural Language Processing (EMNLP), 6327–6340 · Published 1 January 2022 · DOI 10.18653/v1/2022.emnlp-main.425

Open PDFView at the publisherCite this paper

In brief

“Just Fine-tune Twice: Selective Differential Privacy for Large Language Models” is a 2022 conference paper by Weiyan Shi and 5 co-authors, published in Conference on Empirical Methods in Natural Language Processing (EMNLP). OpenAlex counts 106 citations as of 19 November 2026, which puts it in the top 5% of works published in 2022 by citations. It is open access under a CC BY licence: the full text is free to read at Conference on Empirical Methods in Natural Language Processing (EMNLP).

Citations
106
Most cited in
2025 (38)
Field-weighted impact
5.87
References
31
Authors
6

Abstract

Protecting large language models from privacy leakage is becoming increasingly crucial with their wide adoption in real-world products.Yet applying differential privacy (DP), a canonical notion with provable privacy guarantees for machine learning models, to those models remains challenging due to the trade-off between model utility and privacy loss.Utilizing the fact that sensitive information in language data tends to be sparse, Shi et al. (2021) formalized a DP notion extension called Selective Differential Privacy (SDP) to protect only the sensitive tokens defined by a policy function.However, their algorithm only works for RNN-based models.In this paper, we develop a novel framework, Just Fine-tune Twice (JFT), that achieves SDP for state-of-the-art large transformer-based models.Our method is easy to implement: it first finetunes the model with redacted in-domain data, and then fine-tunes it again with the original in-domain data using a private training mechanism.Furthermore, we study the scenario of imperfect implementation of policy functions that misses sensitive tokens and develop systematic methods to handle it.Experiments show that our method achieves strong utility compared to previous baselines.We also analyze the SDP privacy guarantee empirically with the canary insertion attack 1 .

Cite this paper

Formatted from the catalogue record. Check names, capitals and page numbers against the paper before you submit.

APA 7

Shi, W., Shea, R., Chen, S., Zhang, C., Jia, R., & Zhou, Y. (2022). Just Fine-tune Twice: Selective Differential Privacy for Large Language Models. Conference on Empirical Methods in Natural Language Processing (EMNLP), 6327–6340. https://doi.org/10.18653/v1/2022.emnlp-main.425

MLA 9

Shi, Weiyan, et al. “Just Fine-tune Twice: Selective Differential Privacy for Large Language Models.” Conference on Empirical Methods in Natural Language Processing (EMNLP), 2022, pp. 6327–6340. https://doi.org/10.18653/v1/2022.emnlp-main.425.

Chicago

Shi, Weiyan, Ryan Shea, Si Chen, Chiyuan Zhang, Ruoxi Jia, and Yu Zhou. 2022. “Just Fine-tune Twice: Selective Differential Privacy for Large Language Models.” Conference on Empirical Methods in Natural Language Processing (EMNLP): 6327–6340. https://doi.org/10.18653/v1/2022.emnlp-main.425.

Harvard

Shi, W. et al. (2022) ‘Just Fine-tune Twice: Selective Differential Privacy for Large Language Models’, Conference on Empirical Methods in Natural Language Processing (EMNLP), pp. 6327–6340. doi:10.18653/v1/2022.emnlp-main.425.

IEEE

W. Shi, R. Shea, S. Chen, C. Zhang, R. Jia, and Y. Zhou, “Just Fine-tune Twice: Selective Differential Privacy for Large Language Models,” Conference on Empirical Methods in Natural Language Processing (EMNLP), pp. 6327–6340, 2022, doi: 10.18653/v1/2022.emnlp-main.425.

BibTeX

@inproceedings{shi2022just,
  title     = {{Just Fine-tune Twice: Selective Differential Privacy for Large Language Models}},
  author    = {Shi, Weiyan and Shea, Ryan and Chen, Si and Zhang, Chiyuan and Jia, Ruoxi and Zhou, Yu},
  booktitle = {Conference on Empirical Methods in Natural Language Processing (EMNLP)},
  year      = {2022},
  pages     = {6327--6340},
  doi       = {10.18653/v1/2022.emnlp-main.425},
  url       = {https://doi.org/10.18653/v1/2022.emnlp-main.425}
}

Download for a reference manager: BibTeX (.bib) · RIS for Zotero, Mendeley and EndNote (.ris)

Citations by year

YearCitationsShare of the best year
202233
20231111
20242828
20253838
20262626

Its field-weighted citation impact is 5.87: it has been cited 5.87 times as often as the average work of the same field, type and year (1.00 is the world average).

Topic and keywords

Physical Sciences › Computer Science › Artificial Intelligence › Privacy-Preserving Technologies in Data

  • large language models
  • privacy leakage
  • transformer models
  • policy functions
  • privacy-utility trade-off

Work with this paper

Most-cited papers in Computer Science

All highly cited Computer Science papers

Questions about this paper

How do I cite this paper in APA style?

Shi, W., Shea, R., Chen, S., Zhang, C., Jia, R., & Zhou, Y. (2022). Just Fine-tune Twice: Selective Differential Privacy for Large Language Models. Conference on Empirical Methods in Natural Language Processing (EMNLP), 6327–6340. https://doi.org/10.18653/v1/2022.emnlp-main.425

How many times has this paper been cited?

106 times, according to OpenAlex on 19 November 2026. Google Scholar and Scopus count different sets of sources, so their numbers differ.

Is this paper free to read?

Yes. It is open access at Conference on Empirical Methods in Natural Language Processing (EMNLP): sign in and open the PDF from this page.

Who wrote this paper?

Weiyan Shi, Ryan Shea, Si Chen, Chiyuan Zhang, Ruoxi Jia, Yu Zhou.

What is the DOI of this paper?

10.18653/v1/2022.emnlp-main.425, permanently at https://doi.org/10.18653/v1/2022.emnlp-main.425. Sign in and open it from this page.

Bibliographic data from OpenAlex (CC0), last checked 19 November 2026. ResearchGuru does not host this paper; once you are signed in, the links open it at the publisher or the repository that does.