Why not pooling by max, min, std dev? Why padding by zeros?

Pooling: I am curious why choose either max pooling or average pooling. If pool by max, min and std dev respectively, then stack and triple the number of channels, is it possible to capture more features like the changes in a pooled area?

Padding: The padding is set as all zeros. I wonder how it works if setting as the number next to the padding. Like cutting a picture - the edge of the cutoff looks like the edge of the remaining, rather than blank.

Hey @Helen_Zhou,
Welcome back! As you know that Deep Learning is a field which is actively pursued by researchers. So, you are more than welcome to search if any of these techniques of padding have already been implemented by researchers, and look at the results of their research.

If not, then I believe that it’s a great opportunity for you to implement them yourself, and perhaps you could come up with a new technique of padding that will outshine the existing ones, and we will make sure to include it in the curriculum of DeepLearning.AI so that all the learners can know about your work.

Cheers,
Elemento