Comparing distributions: Kernels estimate good representations, l1 distances give good tests
Read OriginalThis technical article revisits the classic statistical problem of two-sample testing, examining whether two observed datasets are drawn from the same underlying distribution. It explains the use of kernel mean embeddings and Maximum Mean Discrepancy (MMD) to compare distributions, and argues that metrics based on L1 geometry can provide superior testing power compared to traditional kernel methods. The content is based on a NeurIPS 2019 conference paper and delves into the mathematical framework of Integral Probability Metrics and Reproducible Kernel Hilbert Spaces.
Comments
No comments yet
Be the first to share your thoughts!
Browser Extension
Get instant access to AllDevBlogs from your browser