Exchangeability in Neural Network and its Application to Dynamic Pruning

Pu, Yi, Tianlang Chen +2

Modern neural networks (NN) contain an ever-growing number of parameters, substantially increasing the memory and computational cost of inference. Researchers have explored various…