[Paper Accepted] One Paper is Accepted by ISIT 2025!

Abstract

Distributional data emerge in modern signal processing and underscore the computation of optimal transport (OT) maps across multiple probability distributions. Nonetheless, recent studies on neural OT maps predominantly focused on the efficient computation of a single map between a couple of distributions. To answer this challenge, we introduce a novel approach to learn the transport maps for new empirical distributions. Specifically, we employ the transformers architecture to produce embeddings from distributional data of various length; these embeddings are then fed into a hypernetwork to generate neural transport maps. Various numerical experiments were conducted to validate the embeddings and the generated transport maps.

Mingchen Jiang, Peng Xu, Xichen Ye, Xiaohui Chen*, Yun Yang*, Yifan Chen*. “Embedding Empirical Distributions for Computing Optimal Transport Maps”, In IEEE International Symposium on Information Theory (ISIT 2025).