No bad local minima: Data independent training error guarantees for multilayer neural networks

Open in new window