Nonuniform random feature models using derivative information

Konstantin Pieper, Zezhong Zhang, Guannan Zhang

We propose nonuniform data-driven parameter distributions for neural network initialization based on derivative data of the function to be approximated. These parameter distributio…