Treat matrices as functions between vector spaces, with kernels, images, rank, and change of basis. This gives a structural view of linear systems instead of only a calculation method.
