All-Optical Image Identification with Programmable Matrix Transformation

TL;DR

This paper presents a programmable all-optical neural network capable of high-speed image classification using matrix transformations and nonlinear photodetection, achieving high accuracy and potential processing speeds up to 74 trillion FLOPs per second.

Contribution

It introduces a novel all-optical neural network architecture with programmable matrix operations and nonlinear activation, enabling high-speed, high-accuracy image classification on a single platform.

Findings

Achieved high accuracy in classifying handwritten digits, objects, and depth images.

Demonstrated potential processing speeds up to 74T FLOPs/sec.

Implemented programmable matrix transformations using spatial light modulators.

Abstract

An optical neural network is proposed and demonstrated with programmable matrix transformation and nonlinear activation function of photodetection (square-law detection). Based on discrete phase-coherent spatial modes, the dimensionality of programmable optical matrix operations is 30~37, which is implemented by spatial light modulators. With this architecture, all-optical classification tasks of handwritten digits, objects and depth images are performed on the same platform with high accuracy. Due to the parallel nature of matrix multiplication, the processing speed of our proposed architecture is potentially as high as7.4T~74T FLOPs per second (with 10~100GHz detector)