This is the final project of CSC413 by Shujie Deng, Lingfei Li and Saifei Liao.
Deep learning has developed rapidly in recent years, allowing us to see the emergence of many neural nets with astonishing performances. In this paper, we compared the standard convolution structure with separable convolution, batch normalization with group normalization and conducted experiments with AlexNet and VGG-16. Furthermore, we added a BatchNorm layer within separable convolution to evaluate the performance. Empirically, we show that separable convolution perform better than standard convolution, and group normalization perform worse than batch normalization with large batch size.