Revisiting Random Binning Feature: Fast Convergence and Strong Parallelizability

Kernel method has been developed as one of the standard approaches for nonlinear learning, which however, does not scale to large data set due to its quadratic complexity in the number of samples. A n