Preprint · 2021 · Open access

Don't Generate Me: Training Differentially Private Generative Models with Sinkhorn Divergence

Tianshi Cao, Alex Bie, Arash Vahdat, Sanja Fidler, Karsten Kreis

arXiv (Cornell University) · Published 1 November 2021 · DOI 10.48550/arxiv.2111.01177

Open PDFView at the publisherCite this paper

In brief

“Don't Generate Me: Training Differentially Private Generative Models with Sinkhorn Divergence” is a 2021 preprint by Tianshi Cao and 4 co-authors, published in arXiv (Cornell University). OpenAlex counts 60 citations as of 19 November 2026. It is open access: the full text is free to read at arXiv (Cornell University).

Citations
60
Most cited in
2024 (21)
References
60
Authors
5

Abstract

Although machine learning models trained on massive data have led to break-throughs in several areas, their deployment in privacy-sensitive domains remains limited due to restricted access to data. Generative models trained with privacy constraints on private data can sidestep this challenge, providing indirect access to private data instead. We propose DP-Sinkhorn, a novel optimal transport-based generative method for learning data distributions from private data with differential privacy. DP-Sinkhorn minimizes the Sinkhorn divergence, a computationally efficient approximation to the exact optimal transport distance, between the model and data in a differentially private manner and uses a novel technique for control-ling the bias-variance trade-off of gradient estimates. Unlike existing approaches for training differentially private generative models, which are mostly based on generative adversarial networks, we do not rely on adversarial objectives, which are notoriously difficult to optimize, especially in the presence of noise imposed by privacy constraints. Hence, DP-Sinkhorn is easy to train and deploy. Experimentally, we improve upon the state-of-the-art on multiple image modeling benchmarks and show differentially private synthesis of informative RGB images. Project page:https://nv-tlabs.github.io/DP-Sinkhorn.

Cite this paper

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

APA 7

Cao, T., Bie, A., Vahdat, A., Fidler, S., & Kreis, K. (2021). Don't Generate Me: Training Differentially Private Generative Models with Sinkhorn Divergence. arXiv (Cornell University). https://doi.org/10.48550/arxiv.2111.01177

MLA 9

Cao, Tianshi, et al. “Don't Generate Me: Training Differentially Private Generative Models with Sinkhorn Divergence.” arXiv (Cornell University), 2021. https://doi.org/10.48550/arxiv.2111.01177.

Chicago

Cao, Tianshi, Alex Bie, Arash Vahdat, Sanja Fidler, and Karsten Kreis. 2021. “Don't Generate Me: Training Differentially Private Generative Models with Sinkhorn Divergence.” arXiv (Cornell University). https://doi.org/10.48550/arxiv.2111.01177.

Harvard

Cao, T. et al. (2021) ‘Don't Generate Me: Training Differentially Private Generative Models with Sinkhorn Divergence’, arXiv (Cornell University). doi:10.48550/arxiv.2111.01177.

IEEE

T. Cao, A. Bie, A. Vahdat, S. Fidler, and K. Kreis, “Don't Generate Me: Training Differentially Private Generative Models with Sinkhorn Divergence,” arXiv (Cornell University), 2021, doi: 10.48550/arxiv.2111.01177.

BibTeX

@misc{cao2021generate,
  title     = {{Don't Generate Me: Training Differentially Private Generative Models with Sinkhorn Divergence}},
  author    = {Cao, Tianshi and Bie, Alex and Vahdat, Arash and Fidler, Sanja and Kreis, Karsten},
  howpublished = {arXiv (Cornell University)},
  year      = {2021},
  doi       = {10.48550/arxiv.2111.01177},
  url       = {https://doi.org/10.48550/arxiv.2111.01177}
}

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

Citations by year

YearCitationsShare of the best year
202111
202277
202388
20242121
20251414
202699

Topic and keywords

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

  • differential privacy
  • generative models
  • Sinkhorn divergence
  • optimal transport
  • privacy-preserving data synthesis
  • bias-variance trade-off
  • gradient estimation
  • image generation

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?

Cao, T., Bie, A., Vahdat, A., Fidler, S., & Kreis, K. (2021). Don't Generate Me: Training Differentially Private Generative Models with Sinkhorn Divergence. arXiv (Cornell University). https://doi.org/10.48550/arxiv.2111.01177

How many times has this paper been cited?

60 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?

Tianshi Cao, Alex Bie, Arash Vahdat, Sanja Fidler, Karsten Kreis.

What is the DOI of this paper?

10.48550/arxiv.2111.01177, permanently at https://doi.org/10.48550/arxiv.2111.01177. 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.