Mayuri Sridhar

Mayuri Sridhar is a PhD candidate in electrical engineering and computer science who focuses on privacy-preserving computation. As a MathWorks Fellow, Mayuri will expand her work on providing statistical guarantees regarding robustness and security for algorithms in realistic deployed settings. Her recent work focuses on PAC privacy, where she has shown the intrinsic relationship between the stability of algorithms and their potential for privatization. This work provides a tight theoretical characterization of the noise required to privatize any algorithm as a function of its covariance, along with extensive experimental results validating the utility of these algorithms. In particular, she shows that privatizing complex algorithms like SVM or K-Means can be done in a black-box manner with a negligible impact on efficacy. MATLAB’s Parallel Computing Toolbox and Deep Learning Toolbox could enable her to extend her work to more complex models like BERT or even GPT. Mayuri’s work has strong potential to advance privacy-preserving computing with broad applications, from providing privacy guarantees when interacting with large public models like ChatGPT to robust learning guarantees on sensitive medical and financial data.