Coherence Awareness in Diffractive Neural Networks

Abstract

Diffractive neural networks hold great promise for applications requiring intensive computational processing. Considerable attention is focused on diffractive networks for either spatially coherent or spatially incoherent illumination. Here, it is illustrated that, as opposed to imaging systems, in diffractive networks the degree of spatial coherence has a dramatic effect. In particular, it is showed that when the spatial coherence length on the object is comparable to the minimal feature size preserved by the optical system, neither the incoherent nor the coherent extremes serve as acceptable approximations. Importantly, this situation is inherent to many settings involving active illumination, including reflected light microscopy, autonomous vehicles and smartphones. Following this observation, a general framework is proposed for training diffractive networks for any specified degree of spatial and temporal coherence, supporting all types of linear and nonlinear layers. Using this method, networks are numerically optimized for image classification, and the dependence of their performance on the coherence properties of the illumination is thoroughly investigated. The concept of coherence-blind networks is further introduced, enabling networks, which have enhanced resilience to changes in illumination conditions. These findings serve as a steppingstone toward adopting all-optical neural networks in real-world applications, leveraging nothing but natural light.