Implementation of two models in Keras for the classification of the CIFAR-10 dataset.
First model
It's a simple model with three Convolution+Max Pooling+Dropuout layers and two fully-connected layers. It has a validation accuracy of 78%.
Second model
The second model uses augmentation and three different types of normalization:
- Batch normalization
- Kernel regularization
- Dropout
I also use a learning rate scheduler to change the lr in the middle of the training
README WIP