Diagram of Pooling Operations

This diagram illustrates two common pooling operations used in deep learning convolutional neural networks. Using a 4x4 input matrix as an example, a 2x2 sliding window with a stride of 2 divides the matrix into four non-overlapping regions. The upper section demonstrates "Max Pooling," which extracts the maximum value from the four elements in each region to produce a 2x2 downsampled matrix; the lower section demonstrates "Average Pooling," which calculates the average value of the four elements in each region, likewise yielding a 2x2 result matrix. The diagram intuitively demonstrates the core mechanism by which pooling layers reduce computational load and extract key features through the reduction of spatial dimensions.