All-optical machine learning using diffractive deep neural networks

Abstract

Deep learning has been transforming our ability to execute advanced inference tasks using computers. Here we introduce a physical mechanism to perform machine learning by demonstrating an all-optical diffractive deep neural network (DNN) architecture that can implement various functions following the deep learning-based design of passive diffractive layers that work collectively. We created 3D-printed DNNs that implement classification of images of handwritten digits and fashion products, as well as the function of an imaging lens at a terahertz spectrum. Our all-optical deep learning framework can perform, at the speed of light, various complex functions that computer-based neural networks can execute; will find applications in all-optical image analysis, feature detection, and object classification; and will also enable new camera designs and optical components that perform distinctive tasks using DNNs.

Polaritonic Neuromorphic Computing Outperforms Linear Classifiers

“…The "*" indicates that in Ref. [31] the recognition is measured on a preselection of digits, actually giving an absolute efficiency of ≈ 81%. The logistic regression recognition rates on the same training and testing sets used in the experiments are shown as black and green open squares for M, S (91%) and L (90.8%), respectively.…”

Image and Depth Estimation with Mask-Based Lensless Cameras

“…Recently, a number of data-driven methods have been proposed to design custom phase masks and optical elements to estimate depth from a single image [29], [30]. An alloptical diffractive deep neural network is proposed in [31], [32], which can perform pattern recognition tasks such as handwritten digits classification using optical mask layers. Such networks can literally process images at a lightning-fast pace with near-zero energy cost.…”

Multifunctional volumetric meta-optics for color and polarization image sensors

“…Thus, we experimentally validate the fabrication-constrained design using a centimeter-scale analog ( Fig. 3) operating at microwave frequencies in the Ka band (26)(27)(28)(29)(30)(31)(32)(33)(34)(35)(36)(37)(38)(39)(40). This device is designed to maximize the sorting efficiency at 18 wavelengths, equally spaced across the Ka band, and for both orthogonal linear polarizations.…”

“…Beyond imaging, we may tailor each pixel to collect a spectrum of interest, such as chemical fluorescence, for use in remote sensing applications [28]. Finally, this approach could be used to control the scattering based on the spatial mode of incident light for high-NA imaging [29], angular selectivity in photovoltaics [30], or automatic object recognition [31].…”