Optically-Realized Compressive Sensing Neural Networks
As optical sensors are developed under strict size, weight, and power (SWaP) constraints, it is advantageous to sample only the required regions of a scene needed to obtain the desired values. Often, optical hardware and machine learning models for this task are developed separately. In this talk, we will present a novel approach for jointly optimizing the hardware and models. We will also discuss techniques for implementing optical constraints within the neural network with Bayesian neural networks and L0/L1 regularization and current research on using this approach for privacy-preserving imaging. Sandia National Laboratories is a multimission laboratory managed and operated by National Technology & Engineering Solutions of Sandia, LLC, a wholly owned subsidiary of Honeywell International Inc., for the U.S. Department of Energy's National Nuclear Security Administration under contract DE-NA0003525. SAND2020-1389 A.
