A team of electrical and computer engineers from the University of California at Los Angeles (UCLA) has developed a physical artificial neural network. They modeled the device based on the functioning of the human brain. They used AI for training the network using a computer. This new AI device is capable of analyzing large volumes of data and identifying objects at the speed of light. They used a 3D printer to create this device at the UCLA Samueli School of Engineering.
Diffractive Deep Neural Network based UCLA Device
We use various network devices in our everyday lives to identify objects. For example, an automated teller machine consists of a computerized camera to “read” the handwritten dollar amounts when we deposit a cheque. Similarly, the internet search engines can quickly match the images to other similar images available in their databases.
However, such systems usually rely on a piece of equipment to identify the object. They, initially, detect the image using a camera or an optical sensor. Then they process the data by using computing programs to figure out what it is.
Recently, the UCLA team has developed a diffractive deep neural network based AI device to analyze a large amount of data at the speed of light. It uses the light diffracting from the object itself to identify that object in a shorter time period compared to a computer. This AI device does not require any advanced computing programs to process the image of the object. Further, it does not consume any energy to process since it functions on the basis of diffraction of light only.
Functioning of the UCLA Device
The UCLA team developed the artificial neural network using a computer-simulated design. They used a 3D printer to generate very thin, 8 centimeter-square polymer wafers. They kept the surfaces of each wafer uneven so that the light coming from the object would diffract in different directions. The layers look opaque to a human eye. However, the researchers showed that the submillimeter-wavelength terahertz frequencies of light could travel through them during the experiments.
Each layer of the device is composed of tens of thousands of tiny pixels or artificial neurons. The light travels through these tiny pixels of the layers. The series of the pixelated layers function together as an “optical network”. The incoming light from the object diffracts toward a single pixel which is assigned to that type of object. The network deep learns the patterns of the diffracted light from each object that assist in the identification process. In this way, the network identifies the object accurately.
Benefits of the UCLA Device
The UCLA device can implement new technologies to speed up the data-intensive processes like sorting and identifying the objects. For example, this device can be mounted in a driverless car to identify obstacles and signs instantaneously and respond accordingly. The driverless car will be able to “read” the stop sign as soon as the light from the sign hits it. Currently, the car needs to image the object using its camera and then process the image to identify the object accurately.
However, this process is a bit time-consuming.
So, the UCLA device can be installed in the car to reduce the delay in this process. Similarly, this device can be used installed in microscopic imaging devices and other such medical identification equipment. By using this technology, millions of cells can be sorted for identifying the signs of a disease.
The research work of the UCLA team suggests that the new artificial intelligence-based passive device can be used to analyze and process images, data, and objects instantly. They modeled the device intuitively based on the working of the brain to process a large amount of information. This technology can be further scaled up to develop new camera designs and unique optical components. These systems can be further implemented in various applications related to medical technology, robotics, and security.