This project works for my pending paper "CNN Compressing Method". I'm trying to modify kernels in hidden layers with linear representation and propotional fitting, deploying with ResNet-20 Network with Cifar-10 datasets, showed in paper Deep Residual Learning for Image Recognition I have tried two ways of representing, which are seperated in the two branches. However one of them failed when testing after modified. So i'm only pending on the other one.