Nowadays, average pooling is used very rarely, because max pooling will usually outperform average pooling. But anyway let's see what average pooling is doing, which isn't that different from max pooling.

Again, it has the same parameters—the filter size, stride, and padding 0151 but instead of outputting the maximum of the selected values, it outputs their average:

In this case, the average of these values will be 4.5, and, when we move two steps and repeat this, we will finally get the following:

The formula we saw for max pooling, which was borrowed from a convolution formula, also works perfectly for average pooling, so these three parameters can be used to control the output dimensions as we want.