Python Deep Learning Cookbook

Getting started with filters and parameter sharing

Let's introduce the most important part of convolutional networks: convolutional layers. In a convolutional layer, we have blocks that convolve over the input data (like a sliding window). This technique shares parameters for each block in such a way that it can detect a feature within the block across the whole input data. The size of a block is called the kernel size or filter size. Summarized, the convolutional layer extracts local features within the total feature set. This is illustrated with an image as input data in the following figure:

Figure 3.1: An example of a convolutional block ...