Multimodal manifold learning using kernel interpolation along geodesic paths

Ori Katz, Roy R. Lederman, Ronen Talmon

Abstract: Highlights•Multimodal spectral analysis combining manifold learning and Riemannian SPD geometry.•Unsupervised, data-driven method with no assumptions about data modalities.•The method uses the geodesic path between kernel matrices of two aligned datasets.•A new multi-manifold learning algorithm that extracts the latent intrinsic common manifold.•Theoretical analysis and empirical results provide evidence for the usefulness of this approach