Shrinking Massive Neural Networks Used to Model Language

Hidden within massive neural networks, leaner subnetworks can complete the same task more efficiently. The trick is finding those “lucky” subnetworks, dubbed winning lottery tickets.