Convolutional neural networks for image shape data

CNNs allow us to build deep learning models on image data without the need to reformat images into a tabular format. The name of this category of deep learning techniques comes from the concept of convolution, which in deep learning refers to applying a filter to image shape data to produce a secondary image shape feature map (shown in Figure 13.2):

Figure 13.2 – A simple example of applying a predefined convolution filter to a 3x3 image shape data point

When training a deep learning model, for example using PyTorch, a convolution filter or other filters that we will introduce later in this chapter will not be predefined but rather learned through the learning process. Convolution and other filters and processes in CNN modeling let us use the methods under this category of deep learning techniques for different image shape data (as we saw in Figure 13.1).

The application of CNNs is beyond supervised...