Summary
Yingtai Xiao is a research scientist based in Sunnyvale with nine years of experience specializing in differential privacy for machine learning, optimization, and practical applications like advertising measurement and database query systems. He holds a PhD from Penn State and has repeatedly delivered state-of-the-art algorithms—most notably ResidualPlanner, which achieved optimal RMSE for the Matrix Mechanism while improving scalability by 10x and speed by 100x. His work spans theory and systems: from privacy-aware ad attribution at TikTok that cut variance 4x, to secure multiparty training and numerical improvements for private deep learning at JD.com. He also developed frameworks that reduce privacy cost or variance (SM-II and conjugate-gradient Newton methods) and explored shared-mechanism designs that save privacy budget when answering correlated queries. Comfortable moving ideas from proofs to production, he blends strong mathematical foundations (BS in Mathematics) with impactful engineering in industry settings. An underappreciated thread in his work is optimizing algebraic and numerical primitives (e.g., ReLU derivatives, fixed-point simplifications) to make private ML practical at scale.
9 years of coding experience
5 years of employment as a software developer
Doctor of Philosophy - PhD, Computer Science, Doctor of Philosophy - PhD, Computer Science at Penn State University
Bachelor of Science - BS, Mathematics, Bachelor of Science - BS, Mathematics at University of Science and Technology of China