Neural networks will switch to optical chips

Stanford University researchers revealed that the training of artificial neural networks directly on the optical chip leads to more energy efficient fulfillment of complex tasks, such as speech recognition or images.

Artificial neural network is a type of artificial intelligence that uses related units to handle information by a method similar to how our brain does. Optical-oriented devices are of great interest, as they can perform calculations in parallel using less energy than electronic. Researchers have developed an optical chip that trains neural networks in the same way as ordinary computers.

In the new protocol, the laser is served through an optical circuit. When exiting the device, the difference from the expected result is calculated. This information is then used to generate a new light signal, which is sent back through the optical network in the opposite direction. The phase management settings can be changed based on this information, and the process can be repeated until the neural network gives the desired result.

Researchers are planning further optimizing the system and want to use it to solve practical tasks. The general approach they has developed can be used with various architectures of neural networks and for other spheres, such as reconfigurable optics.

Recall that