Sampling Algorithms for Black-Box Generative AI with Applications to Fairness

Abstract:

Many generative AI systems are accessible only through black-box query interfaces, so post-processing is often the only way to control the distribution of attributes in their outputs. In this talk, I will present a sampling-based framework for this problem under fixed query budgets. I will describe how close one can get to a desired target distribution and present efficient algorithms with optimal guarantees. I will also discuss applications to fairness, illustrating how these methods can improve the demographic composition of outputs from biased generators.

Jason Klusowski - Bio

Jason M. Klusowski is an Assistant Professor in the Department of Operations Research and Financial Engineering (ORFE) at Princeton University. Prior to joining Princeton, he was an Assistant Professor in the Department of Statistics at Rutgers University, New Brunswick. He received a PhD in Statistics and Data Science from Yale University. His research develops statistical and algorithmic guarantees for modern large-scale predictive systems. He received a Sloan Research Fellowship in Mathematics, NSF CAREER Award, and Howard B. Wentz, Jr., Junior Faculty Award from the School of Engineering and Applied Science (SEAS) at Princeton. He currently serves as an Associate Editor for Bernoulli.

Jason Klusowski