On the use of deep learning techniques for electromagnetic inverse problems

We present a deep learning approach to capture coherent light propagation through free space and invert it so as to obtain images of phase objects from the scattered intensity. The method seems to generalize well even when trained on specific databases, e.g. of faces.