A convergence analysis of Nesterov’s accelerated gradient method in training deep linear neural networks

rate when the width is near-linear in the depth of the network, where is the number of iterations and is a constant depending on the condition number of the feature matrix. Compared to the