Boosting Deep Neural Network Efficiency with Dual-Module Inference

ICML 2020