Deep Learning Quick Reference

Pooling layers

In addition to convolutional layers, convolutional neural networks often use another type of layer called a pooling layer. Pooling layers are used to reduce the dimensionality of a convolutional network as layers of convolutions are added, which reduces overfitting. They have the added benefit of making the feature detectors somewhat more robust.

Pooling layers divide a matrix into non-overlapping sections, and then typically take the maximum value (in the case of max pooling) in each region. Alternatively, an average can be employed; however, it is rarely used at this time. The following figure illustrates this technique: