Understanding the pooling layer

With just a forward pass from a CNN layer of an image, the size of the two-dimensional output data is likely reduced but is still a substantial size. To reduce the size of the data further, a layer type called a pooling layer is used to aggregate and consolidate the values strategically while still maintaining useful information. Think of this operation as an image-resizing method while maintaining as much information as possible. This layer has no parameters for learning and is mainly added to simply and meaningfully reduce the output data. The pooling layer works by applying a similar sliding window filter process with similar configurations as the convolutional layers but instead of applying a dot product and adding a bias, a type of aggregation is done. The aggregation function can be either maximum aggregation, minimum aggregation, or average aggregation. The layers that apply these aggregations are called max pooling, min pooling, and average pooling...