Abstract
We show that neural networks with an absolute value activation function and with network path norm, network sizes and network weights having logarithmic dependence on can -approximate functions that are analytic on certain regions of .