As we have already mentioned several times, neural networks can do the feature engineering part themselves, and this is especially useful for images. Now we will finally see this in action. For that we will use Convolutional Neural Networks, they are a special kind of neural networks that uses special convolutional layers. They are very well suited for image processing.
In the usual neural networks, the layers are fully connected, meaning that each neuron of a layer is connected to all the neurons from the previous layer. For 28 x 28 images such as digits from MNIST, this is not a big deal, but it starts to be a problem for larger images. Imagine that we need to process images of size 300 x 300; in this case, ...
