Neural Networks and Deep Learning took object recognition to a whole new level.

Artificial Intelligence is a revolutionary technology that has changed the field of robotics on its head. Training a machine to make an informed decision was not even imaginable before the emergence of this technology. The formation of Artificial Neural Networks (ANNs) is one of the most impressive achievements of artificial intelligence. Scientists got inspired by biological neural networks and tried to replicate them in computing systems which gave birth to this amazing tool. Generally, machine learning is used to train such systems as they have no prior information at all. Recently, a team comprising of computer and electrical engineers, from the Samueli School of Engineering at UCLA, have created an ANN which has the ability to process massive volumes of data and identify objects at the actual speed of light.

We are surrounded by a wide range of devices which have a computerized camera to identify objects. All of them rely on their camera or optical sensor to capture the object which is then analyzed to figure out what it actually is. However, the device created at UCLA is much more advanced in recognizing objects because it doesn’t need an image for processing.

The ‘Diffractive Deep Neural Network’ of this device makes use of the light bouncing from the surface of an object to reveal its identity. It improves the efficiency of the recognition process by many times as a regular computer will take this much time in simply imaging the object. Similarly, advanced computing programs are also not needed by this extraordinary device because all the decision-making will be done once its optical sensors detect an object. Last but not the least, no energy is needed to operate this device because all it needs is diffraction of light.

Given the property of this device to analyze large amounts of data, it has a lot of applications in the field of microscopic imaging and medicine. For example, it can scan millions of cells to look for any signs of disease. The fact that it speeds up data-intensive tasks leads us to another benefit of this device which is associated with driverless cars. We know that security is a critical factor with these cars and this invention of the UCLA’s engineers is just the thing we need. This technology will enhance the efficiency of these cars by making them to cease instantaneously. The time taken by the current technology to capture the image and then process it will be reduced significantly. Aydogan Ozcan, the Chancellor’s Professor of Electrical and Computer Engineering at the UCLA who is also the Principal Investigator of this research, said,

“This work opens up fundamentally new opportunities to use an artificial intelligence-based passive device to instantaneously analyze data, images and classify objects. 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.”

Initially, computer simulations were used to design these artificial neural networks. After that, 8 cm² polymer wafers were created by making use of a 3D printer. The light was made to diffract in multiple directions due to the invariability in the surfaces of these wafers. Despite the fact that these layers (have tens of thousands of artificial neurons) may seem opaque to naked eye, light having frequency in terahertz and wavelength in sub-millimeter can pass through them. A lot of these layers give birth to an optical network which determines the flow of incoming light.

This network identifies the object immediately because most of the light coming from an object is diffracted towards a single artificial neuron (pixel) that is designated for that type of object. This process was repeated again and again with different objects to train the network about the pattern of diffracted light for each kind of object. Ozcan described that in the following words:

“This is intuitively like a very complex maze of glass and mirrors. 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.”

The researchers observed that this device can project the image of an object placed in front of it to the other side. In addition to that, this network was found extremely accurate with clothing and handwritten numbers. This ANN can be developed at a much larger scale with more layers having hundreds of millions of artificial neurons. The fact that it is incredibly inexpensive adds to its utility. The device created at UCLA can be replicated in less than $50.