3 ms·Equivalent-accuracy accelerated neural-network training using analogue memory1 points by godelmachine 8y ago