All-Optical Neural Network for Deep Learning

Illustration of the first two-layer, all-optical artificial neural network with nonlinear activation functions. These types of functions are required to perform complex tasks such as pattern recognition. (Source: O. Wang, Peng Cheng Lab.)

Even the most powerful computers are still no match for the human brain when it comes to pattern recognition, risk manage­ment, and other similarly complex tasks. Recent advances in optical neural networks, however, are closing that gap by simu­lating the way neurons respond in the human brain. In a key step toward making large-scale optical neural networks practical, researchers have demons­trated a first-of-its-kind multi­layer all-optical artificial neural network.

Generally, this type of arti­ficial intelligence can tackle complex problems that are impossible with traditional compu­tational approaches, but current designs require extensive compu­tational resources that are both time-consuming and energy intensive. For this reason, there is great interest developing practical optical arti­ficial neural networks, which are faster and consume less power than those based on traditional computers.

The researchers from the Hong Kong University of Science and Tech­nology detail their two-layer all-optical neural network and success­fully apply it to a complex classi­fication task. “Our all-optical scheme could enable a neural network that performs optical parallel compu­tation at the speed of light while consuming little energy,” said Junwei Liu, a member of the research team. “Large-scale, all-optical neural networks could be used for applications ranging from image recog­nition to scientific research.”

In conven­tional hybrid optical neural networks, optical components are typi­cally used for linear operations while nonlinear acti­vation functions – the functions that simulate the way neurons in the human brain respond – are usually implemented elec­tronically because nonlinear optics typically require high-power lasers that are difficult to implement in an optical neural network. To overcome this challenge, the researchers used cold atoms with electro­magnetically induced trans­parency to perform nonlinear functions. “This light-induced effect can be achieved with very weak laser power,” said Shengwang Du, a member of the research team. “Because this effect is based on nonlinear quantum inter­ference, it might be possible to extend our system into a quantum neural network that could solve problems intrac­table by classical methods.”

To confirm the capability and feasibility of the new approach, the researchers constructed a two-layer fully-connected all optical neural network with 16 inputs and two outputs. The researchers used their all-optical network to classify the order and disorder phases of the Ising model, a statis­tical model of magnetism. The results showed that the all-optical neural network was as accurate as a well-trained computer-based neural network. The researchers plan to expand the all-optical approach to large-scale all-optical deep neural networks with complex archi­tectures designed for specific practical appli­cations such as image recognition. This will help demons­trate that the scheme works at larger scales.

“Although our work is a proof-of-principle demons­tration, it shows that it may become possible in the future to develop optical versions of arti­ficial intel­ligence,” said Du. “The next generation of arti­ficial intelli­gence hardware will be intrin­sically much faster and exhibit lower power consumption compared to today’s computer-based arti­ficial intelli­gence,” added Liu. (Source: OSA)

Reference: Y. Zuo et al.: All-optical neural network with nonlinear activation functions, Optica 6, 1132 (2019); DOI: 10.1364/OPTICA.6.001132

Link: Dept. of Physics, Hong Kong University of Science and Technology, Hong Kong, China

Speak Your Mind

*