The next transformations that occur in the CNN are processed in pooling layers. The goal of the pooling layers is to reduce the dimensionality of the feature maps output by the convolution layers (but not their depth) while preserving the spatial variance of the original input data. In other words, the size of the data is reduced in order to reduce computational complexity, memory requirements, and training times while overcoming over fitting so that patterns detected during training can be detected in test data even if their appearance varies. There are various pooling algorithms available, given a specified window size, including the following:
- Max pooling: Takes the maximum value in each window
- Average pooling: Takes the ...
