A research team at the University of California, Los Angeles (UCLA) has created an artificial 3D-printed neural network capable of using light photons to process information at an impressive speed.
Electrical engineer Aydogan Ozcan, the lead author of the study, explained that with automatic learning technology, optical tools, and 3D printing, he and his colleagues have created a system that works with light and, unlike conventional computers, requires no energy source other than an initial light source and a simple detector.
The novelty of this technology is not the use of artificial intelligence (AI) systems, but the use of optical technology and the ability to create a “form” of the artificial neural network using 3D printing. According to Ozcan, this system could imitate the eye of an animal that processes light and images very differently from our eye.
The researcher also suggests that the neural network can be used for microscopy and imaging applications in the medical field when used in the shorter wavelengths of optical microscopy.
In order to build up this recognition system, the researchers used deep learning, in which a computer is supplied with visual or acoustic data and trained to recognize certain patterns for the first time. In this sense, the algorithm creates rules about the proposed data and applies them to create new entries.
Now UCLA engineers have trained their neural network to recognize some types of data, including images of handwritten numbers from 0 to 9 and other objects. In each of the proposed cases, the network created a model based on several layers of pixels, each of which had the ability to transmit light. Each pixel represents an artificial neuron that is connected to other neurons in the same layers.
Instead of adapting the mathematical calculations to each neuron, the researchers said that the optical network has tuned its neurons by changing the phase and amplitude of light in each neuron. Instead of having 1 or 0 as a solution in a neuron, each of them sends or reflects incident light to the next layer.
In this way, AI-based hardware has been steered in a new direction by claiming that it is possible to couple optical neural networks with computers to work simultaneously and share the workload. These components, called Deep Diffractive Neural Networks (D2NN), could be scaled using 3D printing to add additional layers and neurons, which is very important because today’s GPUs consume a lot of power and heat.
Via