Tag
该论文提出垂直切片Wasserstein距离(VSW)，一种用于比较meta-measures（分布的分布）的可扩展度量，证明其为有效度量并具有维度无关的统计收敛保证，在效率和精度上优于现有方法。
This paper introduces a difference-of-convex programming framework in Wasserstein space for optimizing non-convex functionals over probability measures, with explicit decompositions for Maximum Mean Discrepancy and Energy Distance, and proves convergence of the lifted convex-concave procedure.