Gradient Descent Provably Optimizes Over-parameterized Neural Networks

Open in new window