Export Reviews, Discussions, Author Feedback and Meta-Reviews

This paper proposes a new embedding structure for the output of nonlinear dimensionality reduction (NLDR). It borrows the space-time concept from physics and shows that the new structure can encode (or represent) more information than conventional Euclidean space. The idea is pioneering and wonderful. It goes beyond all conventional embedding structures and has theoretical guarantee to accommodate more symmetry in the input space. It brings much flexibility to the NLDR problem.