RQ3

We select 4 state-of-the-art model compression techniques proposed in the last two years (i.e., LAMP, Global, Uniform+ and ERK to see whether our framework can speed up the model compression efficiency. Following the above four works, we use ResNet-101 trained on CIFAR-10 as the target model. We conduct controlled experiments according to whether our framework is used for pre-processing before compression. Specifically, we collect the compression rate, model accuracy and number of iterations under the premise of guaranteeing similar prediction accuracy. 

https://github.com/hnurxn/Deep-Arc/RQ3