Matrix Completion via Nonsmooth Regularization of Fully Connected Neural Networks

Novel nonsmooth regularization improves deep neural network performance for matrix completion.