These layers are periodically inserted into a network to reduce the spatial size (width and height) of current representations, as well as volumes in a specific network stage; this serves to reduce the number of parameters and the computational time of the network. It also monitors overfitting. A pooling layer operates on each depth slice of the input volume independently to resize it spatially.
For example, this technique partitions an input image into a set of squares, and for each of the resulting regions, it returns the maximum value as output.
CNNs also use pooling layers located immediately after the convolutional layers. A pooling layer divides input into regions and selects a single representative value (max-pooling ...