3-5. Polling層

A pooling layer that aggregates the information of the convolution layer is provided.
In the figure below, 2*2 units of the feature map are consolidated into one unit.
Detailed diagram is shown below.
There are various ways to reduce the pooling layer.
If the target area is $z_{\mathrm{uv,k}}^{\mathrm{pl}}$ is as follow: Pooling layer units have no concept of weight or bias. In other words, there is no parameter that defines the model.
In addition, there is no concept of activation functions. The general formula for maximum pooling is shown below.