A1282
Title: Efficient algorithms for large-scale optimal transport problems
Authors: Cheng Meng - Renmin University of China (China) [presenting]
Abstract: Optimal transport methods have become increasingly predominant in machine learning, deep learning, statistics, computer vision, and biomedical research. Despite the wide application, existing methods for solving optimal transport problems may suffer from a substantial computational burden when the sample size is large. The Spar-Sink algorithm is introduced to alleviate the computational burden, which utilizes a novel importance sparsification scheme to accelerate the popular Sinkhorn algorithm. This approach can be effectively applied to the entropic optimal transport problem, unbalanced optimal transport problem, Gromov-Wasserstein distance approximation, Wasserstein barycenter estimation, and generative modelling, among others. The application of echocardiography is considered, which is one of the most promising techniques to display the movements of the myocardium. Experiments demonstrate the approach can effectively identify and visualize cardiac cycles, from which one can diagnose heart failure and arrhythmia. To evaluate the numerical accuracy of cardiac cycle prediction, the task of predicting the end-diastole time point using the end-systole one is considered. Results show Spar-Sink performs as well as the classical Sinkhorn algorithm, requiring significantly less computational time.
