Spotlight: May 7, 2019
To learn well, neural networks usually have to be large, with massive datasets. But what if they don’t actually have to be that big? CSAIL researchers have shown that far smaller subnetworks can be trained to make equally accurate predictions. Full story