Fetching the paper…

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