I'm exploring the matrix representation of strided and dilated convolutions, having understood standard 1D convolutions as Toeplitz matrix-vector multiplications. My queries are:
Strided Convolution: How is strided convolution, which downsamples the input, represented in matrix form? Furthermore, how does this downsampling affect the composition of multiple-strided convolutions in matrix multiplication?
Dilated Convolution: How can dilated convolution be represented in a matrix framework?
Insights or mathematical references on these topics would be greatly appreciated.