Source-linked AI summary
All Optical Neural Network with Nonlinear Activation Functions
Ying Zuo, Bohan Li, Yujun Zhao, Yue Jiang, You-Chiuan Chen, Peng Chen, Gyu-Boong Jo, Junwei Liu, Shengwang Du
TL;DR
The paper addresses the bottleneck of experimentally realizing optical nonlinear activation functions in optical neural networks. It demonstrates all-optical networks using SLMs and Fourier lenses for linear operations and EIT in laser-cooled atoms for nonlinear activation, successfully classifying ordered and disordered phases of an Ising model.
Problem
Experimental realization of nonlinear activation functions remains a bottleneck for extending optical neural networks to practical applications.
Method
The authors implement linear operations with spatial light modulators and Fourier lenses, and nonlinear activation functions with electromagnetically induced transparency in cold atoms.
Results
The all-optical neural network successfully classifies ordered and disordered phases of an Ising model, with output-vector fidelity around 99.8% for a Hankel matrix operation.
Takeaways & Limitations
The demonstrated scheme provides fully optical linear and nonlinear neural-network operations and supports constructing different ANN architectures with parallel computation.
Abstract
from arXiv · showhide
Artificial neural networks (ANNs) have now been widely used for industry applications and also played more important roles in fundamental researches. Although most ANN hardware systems are electronically based, optical implementation is particularly attractive because of its intrinsic parallelism and low energy consumption. Here, we propose and demonstrate fully-functioned all optical neural networks (AONNs), in which linear operations are programmed by spatial light modulators and Fourier lenses, and optical nonlinear activation functions are realized with electromagnetically induced transparency in laser-cooled atoms. Moreover, all the errors from different optical neurons here are independent, thus the AONN could scale up to a larger system size with final error still maintaining in a similar level of a single neuron. We confirm its capability and feasibility in machine learning by successfully classifying the order and disorder phases of a typical statistic Ising model. The demonstrated AONN scheme can be used to construct various ANNs of different architectures with the intrinsic parallel computation at the speed of light.