Statistics and Its Interface
Volume 11 (2018)
Discussion on “Double sparsity kernel learning with automatic variable selection and data extraction”
Pages: 423 – 424
DOSK proposed in  aims to perform both variable selection and data extraction at the same time under the “finite sparsity” assumption. In this short note, we propose two alternative approaches based on random projection and importance sampling without such an assumption. Furthermore, we compare these two methods with DOSK empirically in terms of statistical accuracy and computing efficiency.
data extraction, importance sampling, kernel regression, reproducing kernel Hilbert space, random projection, variable selection
Received 12 March 2018
Published 17 September 2018