The following diagram describes an input to the neural network, which runs through many layers (including Convolutions, Subsampling, and Fully connected), before finally getting to the results:
We create the model with two fully connected layers, decreasing in size, followed by an output layer. As convModel describes, we run 3 x 3, 4 x 4, and 5 x 5 filters in parallel at the first layer. For the following convolutional layers, we run 3 x 3 and 5 x 5 filters with decreasing depths due to RAM limits.
In Keras, we can just stack up layers by adding the desired layers one by one. That's exactly what we'll do here. In convModel, the depths ...
