Okay. Now, we are done with the convolution operation. As a result of the convolution operation, we have some feature maps. But the feature maps are too large in dimension. In order to reduce the dimensions of feature maps, we perform a pooling operation. This reduces the dimensions of the feature maps and keeps only the necessary details so that the amount of computation can be reduced.

For example, to recognize a horse from the image, we need to extract and keep only the features of the horse; we can simply discard unwanted features, such as the background of the image and more. A pooling operation is also called a downsampling or subsampling operation, and it makes the CNN translation invariant. Thus, the pooling layer reduces spatial dimensions by keeping only the important features.

There are different types of pooling operations, including max pooling, average pooling, and sum pooling.

In max pooling, we slide over the filter on the input matrix and simply take the maximum value from the filter window, as shown in the following diagram:

As the name suggests, in average pooling, we take the average value of the input matrix within the filter window, and in sum pooling, we sum all the values of the input matrix within the filter window, as shown in the following diagram: