Researchers demonstrate an all-optical neural network for deep learning

Even the most powerful computers are no match for the human brain when it comes to pattern recognition, risk management and other such complex tasks. Recent advances in optical neural networks, however, fill this gap by simulating the way neurons respond to the human brain.

In a key step towards the implementation of large scale optical neural networks, the researchers presented a network of multi-layered all-optical artificial neurons, the first of its kind. In general, this type of artificial intelligence can solve complex problems impossible with traditional computer-based approaches, but current designs require considerable computing resources, which take both time and effort. ;energy. For this reason, there is great interest in developing artificial optical neural networks that are practical, faster and more energy efficient than those based on traditional computers.

In Optica, A journal of the Optical Society specializing in high – impact research, researchers from the Hong Kong University of Science and Technology, Hong Kong, detail their two – layer optical all – optical neural network and the. successfully apply to a complex classification task.

"Our fully optical system could create a network of neurons performing optical computation parallel to the speed of light while consuming little energy," said Junwei Liu, a member of the research team. "Full-scale, all-optical neural networks could be used for applications ranging from image recognition to scientific research."

Build an all-optical network

In conventional hybrid optical neural networks, optical components are typically used for linear operations, whereas nonlinear activation functions – functions that simulate how human brain neurons respond – are typically implemented electronically, and the Nonlinear optics generally require high power lasers that are difficult to use. implement in a network of optical neurons.

To overcome this challenge, researchers used cold atoms with electromagnetically induced transparency to perform nonlinear functions. "This light-induced effect can be achieved with very low laser power," said Shengwang Du, a member of the research team. "Since this effect is based on non-linear quantum interference, it may be possible to extend our system to a quantum neuron network capable of solving insoluble problems by conventional methods."

To confirm the capacity and feasibility of the new approach, the researchers constructed a fully connected two-layer 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 neural network.

Optical neural networks on a larger scale

The researchers plan to extend 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 system works on a larger scale.

"Although our work is a proof of principle, it shows that it might become possible in the future to develop optical versions of artificial intelligence," Du said. "The next generation of artificial intelligence hardware will inherently be much faster and will display lower power consumption than computer-aided artificial intelligence," Liu added.