Adaptive Proximal Gradient Methods for Structured Neural Networks