Convex optimization learning of faithful Euclidean distance representations in nonlinear dimensionality reduction