Definition
Pooling functions reduces the spatial dimensions (width and height) of the input volume while preserving the most important features.
Max Pooling
Selects the maximum value from the region covered by the filter. Good for detecting sharp features like edges. Example: From values [1, 3, 2, 4], max pooling returns 4.
Average Pooling
Calculates the average of all values in the region covered by the filter. Better for preserving background features.
Example: From values [1, 3, 2, 4], average pooling returns 2.5
Use case
- Reduce computation by decreasing dimensions
- Extract dominant features
- Provide some translation invariance
- Reduce overfitting