For implementing the CNN in Gnuastro, we need so many functions. Fortunately, many of these functions are available in Gnuastro. But there are some options in some operations like Convolution that have not been considered in Gnuastro.
In machine learning, There are three main types of padding: same padding, valid padding, and causal padding.
Also, We can determine the stride of the kernels that scroll over the whole of the input. Therefore, If I used the same padding and the stride of 2, we have to determine 2 layers of zero beside the row and column of the input.

In Gnuastro, there is no way to specify the stride and the type of padding in the convolution operation. So we must add these options so that the user of machine learning can select every desired option.