A new method of AI optimization could save companies energy without changing hardware. Developed at the University of Michigan, the open-source optimization framework studies deep learning models during training, pinpointing the best tradeoff between energy consumption and training speed.
