The concept of deep learning is currently being harnessed furiously, enabling computer systems to understand and learn from volumes of data sets, identifying patterns that can enable detection of new data. Now, a group of University of California researchers have come up with a first-of-its-kind neural network that can function on light rather than electricity in order to make conclusion.
The results of the report have been recently published in the journal Science, wherein the researchers have explained their novel working device, their true idea, the performance of current prototype, and applications that they believe can be served by the new neural network. The researchers have created a deep learning network that runs on what is called as D2NN or diffractive deep neural network.
For this, the Researchers constructed tiny plastic plates that were printed with a 3D printer, and each were subjected to a layer of virtual neurons. Since different neurons have the ability to behave distinctly based on its biological counterparts, reflecting or transmitting inbound light. The prototype contains a sensor at the end in order to detect light and interpret the information.
Challenging their own idea, the researchers opted to develop a real neural network that was able to identify numeric values up to nine from zero. As a test, the new system was displayed 55,000 random images with numbers and came up with positive results up to 95 percent. Now that the concept has been proven, the next step in the study is to develop dedicated devices can be commercialized, such as identifying faces from a moving crowd of people.