An unfeasibility view of neural network learning

Authors: Joos Heintz, Luis Miguel Pardo, Enrique Carlos Segura, Hvara Ocar, Andrés Rojas Paredes

Abstract:
We define the notion of a continuously differentiable perfect learning algorithm for multilayer neural network architectures and show that such algorithms do not exist provided that the length of the data set exceeds the number of involved parameters and the activation functions are logistic, tanh or sin.

More information: https://www.sciencedirect.com/science/article/abs/pii/S0885064X22000759