This Jupyter notebook is using keras to build the residual or dense block, and compare with four models in cifar100 classification as following.

• 18 layers residual network with ReLU and batch normalization
• 18 layers plain network with ReLU and batch normalization
• 18 layers residual network with SELU
• 18 layers plain network with SELU

All model was trained with 100 epochs and Adam(learning rate:$10^{-5}$)