Wasserstein Distance in Deep Learning

Abstract

Wasserstein distance, also known as earth mover's distance, has emerged as a powerful distance metric in machine learning for measuring the dissimilarity between two probability distributions. In recent years, Wasserstein distance has been increasingly used in deep learning due to its ability to capture non-local and high-dimensional structure, and its ability to overcome limitations of traditional distance metrics such as Euclidean distance. This survey paper provides an overview of the applications of Wasserstein distance in deep learning, including image generation, domain adaptation, and representation learning, and discusses various techniques for estimating Wasserstein distance in deep learning. Additionally, future research directions and potential applications of Wasserstein distance in emerging areas of deep learning are also discussed. The aim of this survey paper is to provide a comprehensive understanding of Wasserstein distance in deep learning and its potential impact on the future of machine learning.