Title: Curvature correction for distance-based learning on shape space
Authors: Philipp Harms - University of Freiburg (Germany) [presenting]
Abstract: As statistics on manifolds can be prohibitively slow, linearizations are often used in practice to obtain approximations at reasonable computational cost. These approximations are, however, distorted by the curvature of the space. We introduce curvature corrections for distance-based algorithms such as multi-dimensional scaling or agglomerative clustering and show some applications to shape analysis on landmark spaces.