Why is max pooling preferred over average pooling in convolutional neural networks?

Asked Nov 08 '22 at 08:29

Active Nov 08 '22 at 08:29

Viewed 158 times

In most CNN guides max pooling is explained and it is just mentioned that also average pooling exists. Why is max pooling preferred over average pooling? Average (or even median) seems to make more sense to me?

asked Nov 08 '22 at 08:29

lordy

2

Look up translation invariance – Firebug Nov 08 '22 at 08:46
@Firebug , sorry I don't get the connection ... What does taking the maximum value have to do with rotations? https://stats.stackexchange.com/questions/208936/what-is-translation-invariance-in-computer-vision-and-convolutional-neural-netwo – lordy Nov 09 '22 at 15:30
It does not have anything to do with rotations specifically. I mentioned translation invariance – Firebug Nov 11 '22 at 10:24

Why is max pooling preferred over average pooling in convolutional neural networks?

0 Answers0