Imperial joins a 4.2 million euro project to build AI that learns inside physical materials
Most AI runs on power hungry graphics chips. Physicists at Imperial College London want to try something different: letting physical devices do the learning themselves. Their group is part of KAN-DO-IT, a four year European project led by the University of Torino that won 4.2 million euros from the European Innovation Council Pathfinder programme, one of 43 funded out of more than 2,000 bids. It builds on the team’s work with Kolmogorov Arnold Networks made from tiny silicon devices, which shrank networks by up to 600 times, with projected energy savings of up to a thousand times for suitable tasks. The goal is AI that works where power and computing are scarce.