An unfeasability view of neural network learning

Joos Heintz, Hvara Ocar, Luis Miguel Pardo +2

We define the notion of a continuously differentiable perfect learning algorithm for multilayer neural network architectures and show that such algorithms don't exist provided that…