[nurbsnn]

On the contrary, there was a mathematical proof that one-hidden-layer neural network with nonlinearity is enough to represent any function. Using more than 1 hidden seemed a waste.

https://en.wikipedia.org/wiki/Universal_approximation_theore...