Differential privacy provides mathematical guarantees of privacy and is becoming increasingly important as fields ranging from machine learning to healthcare rely more heavily on sensitive data. A central challenge in differentially private linear regression is the high sensitivity of standard estimators, where small changes in the dataset can significantly affect the output. This sensitivity requires the addition of noise to ensure privacy, which in turn increases sample complexity and makes achieving strong privacy guarantees more computationally demanding than in the non-private setting. My project will focus on improving methods for differentially private linear regression, building toward more efficient algorithms with reduced overhead. I also aim to extend these techniques to logistic regression, which is both practically important and theoretically challenging due to its nonlinear objective, which introduces additional difficulties in controlling sensitivity and designing effective noise mechanisms.

I am excited to participate in SuperUROP to apply what I have learned in statistics and machine learning courses to this growing area of research. This project connects theory with challenges arising from real-world data, and I hope to develop a deeper understanding of this intersection through collaboration with Professor Sam Hopkins and Ph.D. student Ittai Rubinstein.