With recent advances in technology, it has become common to see one device or another quickly and accurately identify objects. However, a team of researchers from the University of California, Los Angeles has gone several steps further than creating a device that merely detects and recognizes these items, as their newly developed artificial intelligence device is reportedly capable of identifying objects at the speed of light.
A press release published earlier in the week on the UCLA website detailed the new discovery, a “diffractive deep neural network” which was created with a 3D printer at the university’s Samueli School of Engineering. Unlike existing devices that use a camera, optical sensor, or other types of equipment to image the object, the new system utilizes the light that bounces from the object for fast, accurate identification.
The AI device also differs from earlier systems as it doesn’t use any energy due to its use of light diffraction and doesn’t require any advanced software to further analyze the object’s image and complete the identification process. Furthermore, the components used in the device’s creation are affordable, as a replica version could be created for less than $50.
The UCLA press release explained the methodologies used when the engineers created and tested the AI network. After making a computer-simulated design for the device, the researchers used a 3D printer to craft tiny, slim polymer wafers with uneven surfaces that allow the device to diffract light from an object in multiple directions. Each of the wafers is composed of “tens of thousands” of artificial neurons, with the neurons serving as pixels where the light can pass through in submillimeter-wavelength terahertz frequencies.
In order to improve the network’s accuracy, the researchers used the artificial intelligence technique known as deep learning, which makes a machine “smarter” through repetition of certain patterns. In this case, the network was made to learn the pattern of diffracted light produced by various objects as the light made its way through the AI device. This resulted in the accurate recognition of handwritten numbers and pieces of clothing, the press release noted.
“This is intuitively like a very complex maze of glass and mirrors,” read a statement from the UCLA Chancellor’s Professor of Electrical and Computer Engineering and study principal investigator Aydogan Ozcan.
“The light enters a diffractive network and bounces around the maze until it exits. The system determines what the object is by where most of the light ends up exiting.”
— TwinyBots (@Twinybots) August 5, 2018
According to the researchers behind the new study, the creation of a device that could recognize objects at the speed of light could lead to further advances based on the system. This could include creating smart driving technology that allows a self-driving car to react “instantaneously” after recognizing a stop sign. If the new system is used, such vehicles would identify the stop sign once the light from the sign makes contact with it, instead of waiting until the car’s camera and software respectively image and analyze the object and identify it as a stop sign.
“This work opens up fundamentally new opportunities to use an artificial intelligence-based passive device to instantaneously analyze data, images and classify objects,” said Ozcan.
“This optical artificial neural network device is intuitively modeled on how the brain processes information. It could be scaled up to enable new camera designs and unique optical components that work passively in medical technologies, robotics, security or any application where image and video data are essential.”
Although the AI device created by the UCLA team was rather small, the fact that its components were crafted with a 3D printer could allow for the creation of larger, similar devices with more layers and potentially “hundreds of millions” of artificial neurons. Such devices would potentially be able to identify more objects simultaneously or utilize more intricate techniques when recognizing these objects.
The study detailing the development of the new AI device was first published on July 26 in the journal Science.