Preprint · 2022 · Open access

Differentially Private Decoding in Large Language Models

Jimit Majmudar, Dupuy, Christophe, Charith Peris, Sami Smaili, Rahul Gupta, Richard S. Zemel

arXiv (Cornell University) · Published 26 May 2022 · DOI 10.48550/arxiv.2205.13621

Open PDFView at the publisherCite this paper

In brief

“Differentially Private Decoding in Large Language Models” is a 2022 preprint by Jimit Majmudar and 5 co-authors, published in arXiv (Cornell University). OpenAlex counts 80 citations as of 19 November 2026. It is open access: the full text is free to read at arXiv (Cornell University).

Citations
80
Most cited in
2025 (31)
References
21
Authors
6

Abstract

Recent large-scale natural language processing (NLP) systems use a pre-trained Large Language Model (LLM) on massive and diverse corpora as a headstart. In practice, the pre-trained model is adapted to a wide array of tasks via fine-tuning on task-specific datasets. LLMs, while effective, have been shown to memorize instances of training data thereby potentially revealing private information processed during pre-training. The potential leakage might further propagate to the downstream tasks for which LLMs are fine-tuned. On the other hand, privacy-preserving algorithms usually involve retraining from scratch, which is prohibitively expensive for LLMs. In this work, we propose a simple, easy to interpret, and computationally lightweight perturbation mechanism to be applied to an already trained model at the decoding stage. Our perturbation mechanism is model-agnostic and can be used in conjunction with any LLM. We provide theoretical analysis showing that the proposed mechanism is differentially private, and experimental results showing a privacy-utility trade-off.

Cite this paper

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

APA 7

Majmudar, J., Christophe, D., Peris, C., Smaili, S., Gupta, R., & Zemel, R. S. (2022). Differentially Private Decoding in Large Language Models. arXiv (Cornell University). https://doi.org/10.48550/arxiv.2205.13621

MLA 9

Majmudar, Jimit, et al. “Differentially Private Decoding in Large Language Models.” arXiv (Cornell University), 2022. https://doi.org/10.48550/arxiv.2205.13621.

Chicago

Majmudar, Jimit, Dupuy, Christophe, Charith Peris, Sami Smaili, Rahul Gupta, and Richard S. Zemel. 2022. “Differentially Private Decoding in Large Language Models.” arXiv (Cornell University). https://doi.org/10.48550/arxiv.2205.13621.

Harvard

Majmudar, J. et al. (2022) ‘Differentially Private Decoding in Large Language Models’, arXiv (Cornell University). doi:10.48550/arxiv.2205.13621.

IEEE

J. Majmudar, D. Christophe, C. Peris, S. Smaili, R. Gupta, and R. S. Zemel, “Differentially Private Decoding in Large Language Models,” arXiv (Cornell University), 2022, doi: 10.48550/arxiv.2205.13621.

BibTeX

@misc{majmudar2022differentially,
  title     = {{Differentially Private Decoding in Large Language Models}},
  author    = {Majmudar, Jimit and Christophe, Dupuy, and Peris, Charith and Smaili, Sami and Gupta, Rahul and Zemel, Richard S.},
  howpublished = {arXiv (Cornell University)},
  year      = {2022},
  doi       = {10.48550/arxiv.2205.13621},
  url       = {https://doi.org/10.48550/arxiv.2205.13621}
}

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

Citations by year

YearCitationsShare of the best year
202211
202388
20242323
20253131
20261717

Topic and keywords

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

  • differential privacy
  • large language models
  • privacy-utility trade-off
  • memorization
  • data leakage
  • fine-tuning
  • inference privacy

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?

Majmudar, J., Christophe, D., Peris, C., Smaili, S., Gupta, R., & Zemel, R. S. (2022). Differentially Private Decoding in Large Language Models. arXiv (Cornell University). https://doi.org/10.48550/arxiv.2205.13621

How many times has this paper been cited?

80 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 arXiv (Cornell University): sign in and open the PDF from this page.

Who wrote this paper?

Jimit Majmudar, Dupuy, Christophe, Charith Peris, Sami Smaili, Rahul Gupta, Richard S. Zemel.

What is the DOI of this paper?

10.48550/arxiv.2205.13621, permanently at https://doi.org/10.48550/arxiv.2205.13621. 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.