H.B. Keller Colloquim
This talk will survey some advances in the use of optimal transport for machine learning problems. Optimal transport considers the geometrical properties of transformations of probability distributions, making it a suitable framework for many applications in sampling, generative modeling, fairness, and causal inference. We will study estimators for this problem, characterizing their finite-sample behavior and obtaining distributional limits suitable for practical inference. Additionally, we will explore structural assumptions that improve the statistical and computational performance of these estimators in high dimensions.