Deep optics

The last decades have witnessed the tremendous advance of the deep learning (DL) technic, which has profoundly influenced the optics field. Previous chapters of this book primarily focus on the implementation of neural networks with photonics-based devices. On the opposite side, the emerging DL methods and progressive computational hardware have also unlocked a new paradigm of optics. In this chapter, we attempt to provide a broad overview of how DL fueled the optical field, especially computational optics. A fresh look at optics from the perspective of inverse problems is taken. DL approaches for optical inverse problem solving are discussed, including their benefits, commonly used methods, and inspiration on computational sensing. Two typical examples, that is, lensless imaging and computer-generated holography, are then given in detail. We also posed several related pending questions and future outlooks.