BriefGPT.xyz
May, 2018
使用Wasserstein距离的Sinkhorn逼近的微分特性
Differential Properties of Sinkhorn Approximation for Learning with Wasserstein Distance
HTML
PDF
Giulia Luise, Alessandro Rudi, Massimiliano Pontil, Carlo Ciliberto
TL;DR
应用最优输运及熵正则化计算Wasserstein距离中的Sinkhorn近似算法的梯度，可以提高学习和优化问题的效率，同时通过高阶平滑性，也可以提供统计保证。
Abstract
Applications of
optimal transport
have recently gained remarkable attention thanks to the computational advantages of
entropic regularization
. However, in most situations the Sinkhorn approximation of the Wassers
→