Pooling

Pooling summarizes nearby values.

Max pooling takes the largest value in a local region. Average pooling takes the average.

For values [1, 5, 2, 4], max pooling over the whole group gives:

max = 5

Average pooling gives:

(1 + 5 + 2 + 4) / 4 = 3

Pooling was historically common in CNNs because it reduced spatial size and made features less sensitive to small shifts. Modern architectures use pooling in different ways, and sometimes replace it with strided convolutions or other designs.

The concept remains useful: pooling is local summarization.