A1074
Title: A rolled Gaussian process model for curves on manifolds
Authors: Simon Preston - University of Nottingham (United Kingdom)
Karthik Bharath - University of Nottingham (United Kingdom)
Pablo Lopez-Custodio - Nottingham Trent University (United Kingdom)
Alfred Kume - University of Kent (United Kingdom) [presenting]
Abstract: Curves on manifolds arise as data in various applications, but there are few available statistical models suited to them. The main obstacle is the nonlinearity of manifolds, which makes it difficult to specify useful models. Hence, one strategy is to flatten the manifold and then exploit the flattened space for modelling, but how the flattening is done is crucial because it induces distortions. "Unrolling" and "unwrapping" are harnessed to flatten the manifold in a particular optimal way that minimizes distortion local to the data, opening the path to a class of random effect models that are easy to work with.
