Text this: Nonsmooth Optimization-Based Hyperparameter-Free Neural Networks for Large-Scale Regression