Reviews: Learning Kernels with Random Features

The proposed approach is very interesting and novel. The idea of using random features to speed up kernel alignment is brilliant. The paper is well written and organized in a principled way. The authors provide a theoretical analysis of the method, guaranteeing consistency of the learned kernel and generalization properties of the resulting estimator. The experimental part allows to better visualize the method, shows how it can be used to identify sparse features in high dimensions and compares the accuracy and computational time of the proposed method on 3 benchmark datasets.