<ul><li>Introduction of sliced optimal transport dataset distance (s-OTDD) for dataset comparison without the need for training.</li><li>Moment Transform Projection (MTP) is utilized to map labels to real numbers, transforming datasets into one-dimensional distributions.</li><li>s-OTDD is defined as the expected Wasserstein distance between the projected distributions, achieving (near-)linear computational complexity and independence from the number of classes.</li><li>s-OTDD shows correlation with optimal transport dataset distance, efficiency compared to other discrepancy measures, and performance indicators in transfer learning and data augmentation.</li></ul>

Lightspeed Geometric Dataset Distance via Sliced Optimal Transport

Discover more