Selecting Optimal Subset to release under Differentially Private M-estimators from Hybrid Datasets

Area of Science:

• Data privacy
• Statistical learning
• Health informatics

Background:

• Growing concerns regarding health data sharing and privacy.
• Need for methods to leverage public data for research without compromising private datasets.

Purpose of the Study:

• To develop a method for selecting optimal public data subsets for M-estimators within differential privacy (DP).
• To address the challenge of utilizing public information to enhance understanding of private datasets securely.

Main Methods:

• Constructing weighted private density estimation from hybrid datasets under DP.
• Analyzing the accuracy of DP M-estimators using hybrid datasets.
• Developing an algorithm for optimal public data subset selection under DP.

Main Results:

• The bias-variance tradeoff in M-estimator performance is characterized by the sample size of the released dataset.
• An algorithm was developed to identify the optimal public data subset for release under DP.

Conclusions:

• The findings provide a guideline for real-world applications of privacy-preserving data analysis.
• Simulation studies and real-data applications validate the proposed method for optimal subset selection.