推土机距离的正定性及其与集合交集的关系
机器学习
2018-08-22 v3 机器学习
摘要
正定核是机器学习中的重要工具,它通过隐式线性化问题几何结构,使得原本困难或难解的问题得到有效解决。本文中,我们发展了推土机距离(Earth Mover's Distance, EMD)的集合论解释,并提出了推土机交集(Earth Mover's Intersection, EMI)——一种针对大小不同集合的EMD正定类比。我们给出了EMD或某些EMD近似为负定的条件。我们还提出了一种保正定变换,可应用于任意核,并可用于推导基于EMD的正定核,且表明Jaccard指数仅是此变换的结果。最后,我们在多种计算机视觉任务中评估了基于EMI和所提变换的核与EMD的对比,表明即使使用不定核技术,EMD通常也处于劣势。
引用
@article{arxiv.1510.02833,
title = {On the Definiteness of Earth Mover's Distance and Its Relation to Set Intersection},
author = {Andrew Gardner and Christian A. Duncan and Jinko Kanno and Rastko R. Selmic},
journal= {arXiv preprint arXiv:1510.02833},
year = {2018}
}
备注
Major revision based on referee comments. Includes significant reorganization of content, new title, new propositions, revised proofs of previous propositions, and additional experiments with new data, kernels, and indefinite kernel techniques