Advanced Algorithm for Energy-Efficient Neural Networks

EPFL researchers have developed a groundbreaking algorithm that efficiently trains analog neural networks, offering an energy-efficient alternative to traditional digital networks. This method, which aligns more closely with human learning, has shown promising results in wave-based physical systems and aims to reduce the environmental impact of deep neural networks. (AI-generated DALL-E 3 conceptual image depicting light waves passing through a physical system.) Credit: © LWE/EPFL