Photonic computing using reconfigurable liquid crystal based scattering media

We present an optical learning framework based on reconfigurable liquid crystal/polymer composite (LC/PC) scattering media. Our design leverages multiple scatterings between the LC/PC sample and a spatial light modulator (or a digital mirror device) that encodes the input data to realize random nonlinear optical mapping in a tunable manner. The system is applied to several learning tasks to demonstrate its capabilities. The reconfigurability of the LC/PC can optimize the learning performance, and enable photonic ensemble learning, which further improves the overall performance.