This network gets its name from the convolutional operator. In layman’s terms, the mathematical combination (Dot Product) of two functions. In CNNs this this operation is implemented in what is called a feature detector, feature, or—most commonly—a kernel.
A kernel is a mini matrix (orders of magnitude?) smaller than the input. The kernel slides across the image, taking the dot product of the two matrices and saves the result to a new image called the “feature map”.
You can then pass the feature layer into a pooling layer to shrink it and save computation
You can then repeat this process on the feature maps to get more complex layers
This video has a wonderful visualization.
10/12/25