Wasserstein-2 benchmark
收藏arXiv2025-09-30 收录
下载链接：
http://github.com/facebookresearch/w2ot
下载链接
链接失效反馈官方服务：
资源简介：
该数据集是一个Wasserstein-2基准测试，它为学习欧几里得Wasserstein-2最优传输的势能提供了一个具体且已建立基准的评价环境。该基准测试包含了已知的（真实）最优传输映射，并允许通过未解释方差百分比（ℒ²-UVP）指标来评估学习到的传输映射。该任务涉及在分布之间的最优传输。
This dataset is a Wasserstein-2 benchmark that provides a concrete and well-established evaluation environment for learning the potentials of Euclidean Wasserstein-2 optimal transport. This benchmark includes known (ground-truth) optimal transport maps, and enables evaluation of learned transport maps using the Percentage of Unexplained Variance (ℒ²-UVP) metric. This task focuses on optimal transport between probability distributions.
