Journal article · 2024 · Open access

Differentially Private Fine-tuning of Language Models

Da Yu, Saurabh Naik, Artūrs Bačkurs, Sivakanth Gopi, Huseyin A. Inan, Gautam Kamath, Janardhan Kulkarni, Yin Tat Lee, Andre Manoel, Lukas Wutschitz, Sergey Yekhanin, Huishuai Zhang

Journal of Privacy and Confidentiality, 14(2) · Published 24 June 2024 · DOI 10.29012/jpc.880

Open full textCite this paper

In brief

“Differentially Private Fine-tuning of Language Models” is a 2024 journal article by Da Yu and 11 co-authors, published in Journal of Privacy and Confidentiality. OpenAlex counts 323 citations as of 19 November 2026, which puts it in the top 6% of works published in 2024 by citations. It is open access under a CC BY-NC-ND licence: the full text is free to read at Journal of Privacy and Confidentiality.

Citations
323
Most cited in
2025 (94)
Field-weighted impact
19.68
References
97
Authors
12

Abstract

We give simpler, sparser, and faster algorithms for differentially private fine-tuning of large-scale pre-trained language models, which achieve the state-of-the-art privacy versus utility tradeoffs on many standard NLP tasks. We propose a meta-framework for this problem, inspired by the recent success of highly parameter-efficient methods for fine-tuning. Our experiments show that differentially private adaptations of these approaches outperform previous private algorithms in three important dimensions: utility, privacy, and the computational and memory cost of private training. On many commonly studied datasets, the utility of private models approaches that of non-private models. For example, on the MNLI dataset we achieve an accuracy of $87.8\%$ using RoBERTa-Large and $83.5\%$ using RoBERTa-Base with a privacy budget of $\epsilon = 6.7$. In comparison, absent privacy constraints, RoBERTa-Large achieves an accuracy of $90.2\%$. Our findings are similar for natural language generation tasks. Privately fine-tuning with DART, GPT-2-Small, GPT-2-Medium, GPT-2-Large, and GPT-2-XL achieve BLEU scores of 38.5, 42.0, 43.1, and 43.8 respectively (privacy budget of $\epsilon = 6.8,\delta=$ 1e-5) whereas the non-private baseline is $48.1$. All our experiments suggest that larger models are better suited for private fine-tuning: while they are well known to achieve superior accuracy non-privately, we find that they also better maintain their accuracy when privacy is introduced.

Cite this paper

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

APA 7

Yu, D., Naik, S., Bačkurs, A., Gopi, S., Inan, H. A., Kamath, G., Kulkarni, J., Lee, Y. T., Manoel, A., Wutschitz, L., Yekhanin, S., & Zhang, H. (2024). Differentially Private Fine-tuning of Language Models. Journal of Privacy and Confidentiality, 14(2). https://doi.org/10.29012/jpc.880

MLA 9

Yu, Da, et al. “Differentially Private Fine-tuning of Language Models.” Journal of Privacy and Confidentiality, vol. 14, no. 2, 2024. https://doi.org/10.29012/jpc.880.

Chicago

Yu, Da, Saurabh Naik, Artūrs Bačkurs, Sivakanth Gopi, Huseyin A. Inan, Gautam Kamath, Janardhan Kulkarni, et al. 2024. “Differentially Private Fine-tuning of Language Models.” Journal of Privacy and Confidentiality 14 (2). https://doi.org/10.29012/jpc.880.

Harvard

Yu, D. et al. (2024) ‘Differentially Private Fine-tuning of Language Models’, Journal of Privacy and Confidentiality, 14(2). doi:10.29012/jpc.880.

IEEE

D. Yu et al., “Differentially Private Fine-tuning of Language Models,” Journal of Privacy and Confidentiality, vol. 14, no. 2, 2024, doi: 10.29012/jpc.880.

BibTeX

@article{yu2024differentially,
  title     = {{Differentially Private Fine-tuning of Language Models}},
  author    = {Yu, Da and Naik, Saurabh and Bačkurs, Artūrs and Gopi, Sivakanth and Inan, Huseyin A. and Kamath, Gautam and Kulkarni, Janardhan and Lee, Yin Tat and Manoel, Andre and Wutschitz, Lukas and Yekhanin, Sergey and Zhang, Huishuai},
  journal   = {Journal of Privacy and Confidentiality},
  year      = {2024},
  volume    = {14},
  number    = {2},
  doi       = {10.29012/jpc.880},
  url       = {https://doi.org/10.29012/jpc.880}
}

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

Citations by year

YearCitationsShare of the best year
202133
20222929
20235454
20247272
20259494
20267171

Its field-weighted citation impact is 19.68: it has been cited 19.68 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

  • differential privacy
  • large language models
  • parameter-efficient fine-tuning
  • privacy-utility trade-off
  • natural language generation
  • RoBERTa
  • GPT-2
  • meta-framework

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?

Yu, D., Naik, S., Bačkurs, A., Gopi, S., Inan, H. A., Kamath, G., Kulkarni, J., Lee, Y. T., Manoel, A., Wutschitz, L., Yekhanin, S., & Zhang, H. (2024). Differentially Private Fine-tuning of Language Models. Journal of Privacy and Confidentiality, 14(2). https://doi.org/10.29012/jpc.880

How many times has this paper been cited?

323 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 Journal of Privacy and Confidentiality: sign in and open the full text from this page.

Who wrote this paper?

Da Yu, Saurabh Naik, Artūrs Bačkurs, Sivakanth Gopi, Huseyin A. Inan, Gautam Kamath, Janardhan Kulkarni, Yin Tat Lee, Andre Manoel, Lukas Wutschitz, Sergey Yekhanin, Huishuai Zhang.

What is the DOI of this paper?

10.29012/jpc.880, permanently at https://doi.org/10.29012/jpc.880. 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.