What is global average pooling and why is it preferred over fully connected layers in modern CNNs?

Global Average Pooling (GAP) reduces each feature map to a single value, replacing fully connected layers to reduce overfitting and parameters.

Global Average Pooling computes the average of each feature map, converting spatial dimensions into a single value per channel. This removes the need for large fully connected layers, significantly reducing parameters and overfitting risk. It enforces stronger correspondence between feature maps and categories, improving interpretability and generalization.