NeurIPS 2020

Learning Manifold Implicitly via Explicit Heat-Kernel Learning

Meta Review

Four knowledgeable reviewers support acceptance of the paper in view of the strong theoretical analysis it provides for learning the heat kernel on a manifold via Wasserstein Gradient Flow, and its compelling performance in applications to SVGD and deep generative modeling. The paper is therefore accepted, and we ask the authors to implement the changes they proposed to make in their rebuttal before submitting the camera ready version.