Customized Privacy Mechanisms for Statistical Inference

Principal Investigator: Jordan Awan

Differential privacy (DP) is the state-of-the-art framework for formal privacy protection, but many available DP methods are designed primarily for estimation. On the other hand, in many scientific problems, it is important to have a complete statistical analysis, which may include 1) a particular statistical model, 2) estimation, and 3) uncertainty quantification (such as confidence intervals and hypothesis tests). In this project, we design DP mechanisms specifically for these statitistical tasks, focusing primarily on the uncertainty quantification. One general technique we explore is the use of the bootstrap in combination with a privacy mechanism to understand the sampling distribution of the private summaries. Besides general statistical applications, we also study the particular problem of valid causal inference from both randomized and observational studies.

Personnel

Other Faculty: Guang Cheng, Professor of Statistics and Data Science, University of California, Los Angeles Salil Vadhan, Vicky Joseph Professor of Computer Science and Applied Mathematics, Harvard University Aleksandra Slavkovic, Professor of Statistics, Penn State

Students: Zhanyu Wang (graduated) Yuki Ohnishi Yue Wang

Representative Publications

• Ohnishi, Yuki, and Jordan Awan. "Locally Private Causal Inference for Randomized Experiments." arXiv preprint arXiv:2301.01616 (2023).

   

  Awan, Jordan, and Salil Vadhan. "Canonical noise distributions and private hypothesis tests." The Annals of Statistics 51, no. 2 (2023): 547-572.
   
  Wang, Zhanyu, Guang Cheng, and Jordan Awan. "Differentially private bootstrap: New privacy analysis and inference strategies." arXiv preprint arXiv:2210.06140 (2022).
   
  Awan, Jordan, and Yue Wang. "Differentially Private Kolmogorov-Smirnov-Type Tests." arXiv preprint arXiv:2208.06236 (2022).

   

  Awan, Jordan, and Aleksandra Slavković. "Differentially private inference for binomial data." Journal of Privacy and Confidentiality 10, no. 1 (2020): 1-40.

   

  Awan, Jordan, and Aleksandra Slavković. "Differentially private uniformly most powerful tests for binomial data." Advances in Neural Information Processing Systems 31 (2018).

   

Keywords: bootstrap, confidence interval, Differential Privacy, hypothesis test