Michael Kearns

Title: Michael Kearns: Innovator in Synthetic Data Generation

Introduction

Michael Kearns is a prominent inventor based in Philadelphia, PA (US). He has made significant contributions to the field of data privacy and synthetic data generation. With a total of 2 patents to his name, Kearns is recognized for his innovative approaches to complex statistical queries.

Latest Patents

Kearns' latest patents focus on generating relaxed synthetic data using adaptive projection. This algorithm is designed to release answers to a vast number of statistical queries, such as k-way marginals, while adhering to differential privacy. The algorithm operates on a private dataset, utilizing simple perturbation to derive answers and subsequently seeks to create a synthetic dataset that closely aligns with these noisy answers. By employing a continuous relaxation of the synthetic dataset domain, the projection loss becomes differentiable, enabling the application of efficient machine learning optimization techniques. Instead of addressing all queries at once, the algorithm strategically utilizes a privacy budget by iteratively identifying queries with high error in the relaxed synthetic data and repeating the projection process. This method proves effective across various parameters and datasets, particularly when the privacy budget is limited or the query class is extensive.

Career Highlights

Michael Kearns is currently employed at Amazon Technologies, Inc., where he continues to advance his research and development in synthetic data generation. His work is pivotal in enhancing data privacy measures while maintaining the utility of statistical data.

Collaborations

Kearns collaborates with notable colleagues, including Sergul Aydore and William Brown. Their combined expertise contributes to the innovative projects at Amazon Technologies, Inc.

Conclusion

Michael Kearns stands out as a key figure in the realm of synthetic data generation and privacy. His innovative algorithms and collaborative efforts are shaping the future of data privacy and statistical analysis.