Python Deep Learning Cookbook

Applying pooling layers

A popular optimization technique for CNNs is pooling layers. A pooling layer is a method to reduce the number of trainable parameters in a smart way. Two of the most commonly used pooling layers are average pooling and maximum (max) pooling. In the first, for a specified block size the inputs are averaged and extracted. For the latter, the maximum value within a block is extracted. These pooling layers provide a translational invariance. In other words, the exact location of a feature is less relevant. Also, by reducing the number of trainable parameters we limit the complexity of the network, which should prevent overfitting. Another benefit is that it will reduce the training and inference time significantly.

In ...