Researchers at the Massachusetts Institute of Technology produced a new chip for processing neural computing. It is designed for use in devices with limited autonomy, such as smartphones and tablets. With its help, the energy consumption for processing the tasks of machine learning is reduced by 95%.
“The research shows impressive results for the implementation of energy-efficient processing of operations with arrays of memory. This, of course, makes it possible to use more sophisticated neural networks for the classification of images and video in the Internet of things, “said Dario Gil, vice president of the Artificial Intelligence Department of the Massachusetts Institute of Technology.
Neural networks allow you to process huge amounts of data, providing previously unreachable opportunities. True, this requires a lot of processing power and, consequently, energy expended. That is why the possibilities of neural networks in modern smartphones are extremely limited. The new MIT chip increases the speed of neural computing by 3-7 times, consuming much less energy.