Learning on graphs using Orthonormal Representation is Statistically Consistent

Existing research \cite{reg} suggests that embedding graphs on a unit sphere can be beneficial in learning labels on the vertices of a graph. However the choice of optimal embedding remains an open issue.