I tried some network pruning techniques, mainly learned from A Kernel Redundancy Removing Policy for Convolutional Neural Network
I used cifar-10 dataset and tried two network structures: squeezenet(shallow) and resnet(deep)
still need improvement for better performance