pdf
Deep learning theory
Loss surface of neural networks
A Convergence Theory for Deep Learning via Over-Parameterization
On the Convergence Rate of Training Recurrent Neural Networks