juniorCNN
What is pooling in CNN and why is it used?
Updated May 15, 2026
Short answer
Pooling reduces spatial dimensions while retaining important features, improving efficiency and invariance.
Deep explanation
Pooling layers downsample feature maps by summarizing regions (max or average pooling). This reduces computational cost and helps the model become invariant to small translations. Max pooling keeps strongest activations, while average pooling smooths representations.
Real-world example
Reducing image resolution while preserving object detection capability.
Common mistakes
- Overusing pooling leading to loss of spatial detail.
Follow-up questions
- What is max pooling vs average pooling?
- Does pooling have trainable parameters?