Maximum VS Summarization

Question

I _partially understand Max-pooling after reading Convolutional Neural Networks (LeNet) :

Another important CNN concept is maximum unification, which is a form of non-linear trigger sampling. The maximum union divides the input image into a set of non-overlapping rectangles and, for each such subdomain, displays the maximum value.

How about summation? I could not find an article that was easy to understand.

Answer 1

. - . , , , , , .

, . - , .

, . . ( ) .

UPDATE:

/ , , max sum/mean. .