Pooling
Pooling is an important operation that takes play in the pooling layer of a CNN. It is a technique used to downsample the spatial dimension of individual feature maps generated by the convolutional layers. Let's examine some important types of pooling layers. We’ll begin by exploring max pooling, as shown in Figure 7.13. Here, we see how max pooling operations work. The pooling layer simply takes the highest value from each region of the input data.
Figure 7.13 – A max pooling operation
Max pooling enjoys several benefits as it is intuitive and easy to implement. It is also efficient since it simply extracts the highest value in a region, and it has been applied with good effect across diverse tasks.
Average pooling, as the name suggests, reduces the data dimensionality by taking the average value for a designated region, as illustrated in Figure 7.14.
Figure 7.14 – An average pooling operation...