AI on a chip that can recognize images at light speed
Researchers have implemented for the first time a form of AI known as a deep neural networks into a device based on light. They’ve thus come closer to creating a machine which processes what it \”sees\”, very quickly and efficiently, like humans.
The photonic deep neural networks can, for example, classify an image in under 570 picoseconds or almost 2 billion images every second. For context, fluid footage is typically shot at a frame rate between 23 and120 frames per second.
The authors of the University of Pennsylvania wrote: \”Direct, clockless processing of optical information eliminates analog to digital conversion and the need for a large module of memory, allowing for faster and more efficient neural networks for next-generation deep learning systems.\”