Training algorithm breaks barriers to deep physical neural networks

Researchers have developed an algorithm to train an analog neural network just as accurately as a digital one, enabling the development of more efficient alternatives to power-hungry deep learning hardware.