The Wasserstein distance compares two probability distributions. 瓦瑟斯坦距离用于比较两个概率分布。
In generative modeling, optimizing the Wasserstein distance can produce more stable training because it provides a smoother notion of discrepancy between model and data distributions. 在生成建模中,优化瓦瑟斯坦距离往往能让训练更稳定,因为它为模型分布与数据分布之间的差异提供了更平滑的度量方式。