Solving Inverse Problems With Deep Neural Networks – Robustness Included?