#cnn_cifar10 (best viewed as raw file)

A simple CNN implementation using theano and keras. Accuracy 78%

''' input layer - 2D_conv layer with 32 feature map with a size of 33. Followed by dropout to 20% 2nd layer - 2D_conv layer with 32 feature map with a size of 33. Followed by a max pooling layer of size 22 3rd layer - 2D_conv layer with 64 feature map of size 33. Follwed by dropout layer to 20% 4th layer - 2D_conv layer with 64 feature map of size 33. Follwed by a max pooling layer of size 22 5th layer - 2D_conv layer with 128 feature map of size 33. Followed by the same dropout layer set to same 20% 6th layer - 2D_conv layer with 128 feature map of size 33. Followed by a max pooing layer of size 2*2 7th layer - Flatten layer to flatten all the feature and weights. Follwed by dropout layer to 20% 8th layer - A fully connected layer with 1024 features with relu activation Followed by a dropout layer of 20% 9th layer - Again a fully connected layer with 512 features with relu Dropout of 20% output layer - Fully connected output layer with 10 units and a softmax acivation function'''