Training Neural Networks as Learning Data-adaptive Kernels: Provable Representation and Approximation Benefits

Open in new window