Thanks for pointing this reference. I agree that SIGN is overall similar in principle (we are not the first to propose «inception» graph networks as I note in a comment – to the best of my knowledge the first such architecture was done by Anees Kazi). Both Mixhop and SGCN use powers of a diffusion operator. In SIGN we use general multi-hop diffusion operators (for example motif-based) that show some improvement over simple powers.