Conditional Transport

Conditional transport (CT) is a framework leveraging optimal transport theory to align and compare probability distributions, particularly when conditioned on a shared variable. Current research focuses on developing efficient neural network architectures, such as those based on neural ODEs or partially input convex neural networks (PICNNs), to solve CT problems, particularly in high-dimensional spaces. Applications span diverse fields, including zero-shot learning, counterfactual fairness analysis, Bayesian inference, and point cloud registration, where CT offers robust and theoretically grounded methods for handling complex data relationships. The ability to efficiently and accurately estimate conditional transport plans promises significant advancements in various machine learning and data analysis tasks.