What is the role of pooling in a convolutional neural network?
Instruction: Define pooling and discuss its purpose in CNNs.
Updated
Example Answer
Pooling reduces the spatial resolution of a feature map. Max pooling retains the strongest activation in each window; average pooling retains the mean. That can reduce later computation and make a classifier less sensitive to small position changes, but it also discards detail.
A standard pooling layer has no learned weights. It reduces the number of activations, not the convolution's existing parameters. It may reduce parameters in a later fully connected layer because that layer receives a smaller input. I would avoid aggressive pooling when precise locations matter, such as segmenting a small defect, and compare it with strided convolution.
Worked Example
With a 2 × 2 window and stride 2, this single-channel input becomes one value:
1 3 max pooling → 4
2 4 average pooling → 2.5
The two outputs summarize different properties. Neither preserves the original pixel locations or guarantees translation invariance.
Follow-up
For dense prediction, how much spatial detail can the task afford to lose?