2D convolutions are the most common due to their application in images; they are an element-wise multiplication and summation. For example, for the convolution in the diagram below, we have:
Some terminology:
- Kernel: The small matrix sliding over the input data
- Stride: Number of pixels by which the kernel moves as it slides over the input data
- Padding: Extra pixels around the edges of the input data to allow the kernel to fit properly at the borders of the input.
A kernel applied to a input with a padding using strides.