Nearly Optimal VC-Dimension and Pseudo-Dimension Bounds for Deep Neural Network Derivatives | Read Paper on Bytez