Hands-On Artificial Intelligence for IoT

Pooling layer

The convolution layer is followed conventionally by a pooling layer. The purpose of the pooling layer is to progressively reduce the size of the representation, and thus, reduce the number of parameters and computations in the network. Thus, it down samples the information as it propagates through the network in feed forward manner.

Here again, we have a filter, traditionally people prefer a filter of size 2×2, and it moves with a stride of two pixels in both directions. The pooling process replaces the four elements under the 2×2 filter by either the maximum value of the four (Max Pooling) or the average value of the four (Average Pooling). In the following diagram, you can see the result of pooling operation on a 2D single ...