降維(比如PCA或者random projection) -> KDE(kernel density estimation)來(lái)估算密度 -> KL divergence