Learning from uncertain curves: The 2-Wasserstein metric for Gaussian processes

We introduce a novel framework for statistical analysis of populations of non-degenerate Gaussian processes (GPs), which are natural representations of uncertain curves. This allows inherent variation or uncertainty in function-valued data to be properly incorporated in the population analysis. Using the 2-Wasserstein metric we geometrize the space of GPs with L2 mean and covariance functions over compact index spaces. We prove uniqueness of the barycenter of a population of GPs, as well as convergence of the metric and the barycenter of their finite-dimensional counterparts. This justifies practical computations. Finally, we demonstrate our framework through experimental validation on GP datasets representing brain connectivity and climate development. A Matlab library for relevant computations will be published at https://sites.google.com/view/antonmallasto/software.

[1]  J. A. Cuesta-Albertos,et al.  On lower bounds for theL2-Wasserstein metric in a Hilbert space , 1996 .

[2]  L. Ambrosio,et al.  Gradient Flows: In Metric Spaces and in the Space of Probability Measures , 2005 .

[3]  C. Givens,et al.  A class of Wasserstein metrics for probability distributions. , 1984 .

[4]  P. C. Álvarez-Estebana,et al.  Uniqueness and approximate computation of optimal incomplete transportation plans , 2011 .

[5]  Marco Cuturi,et al.  Principal Geodesic Analysis for Probability Measures under the Optimal Transport Metric , 2015, NIPS.

[6]  Asuka Takatsu Wasserstein geometry of Gaussian measures , 2011 .

[7]  Søren Hauberg,et al.  A Random Riemannian Metric for Probabilistic Shortest-Path Tractography , 2015, MICCAI.

[8]  Mehrtash Tafazzoli Harandi,et al.  Approximate infinite-dimensional Region Covariance Descriptors for image classification , 2015, 2015 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP).

[9]  C. Villani Topics in Optimal Transportation , 2003 .

[10]  J. A. Cuesta-Albertos,et al.  A fixed-point approach to barycenters in Wasserstein space , 2015, 1511.05355.

[11]  B. Rajput,et al.  Gaussian measures on Lp spaces, 1 ≤ p < ∞☆ , 1972 .

[12]  Kim C. Border,et al.  Infinite Dimensional Analysis: A Hitchhiker’s Guide , 1994 .

[13]  Guillaume Carlier,et al.  Barycenters in the Wasserstein Space , 2011, SIAM J. Math. Anal..

[14]  Xiao Yang,et al.  Uncertainty Quantification for LDDMM Using a Low-Rank Hessian Approximation , 2015, MICCAI.

[15]  M. Gelbrich On a Formula for the L2 Wasserstein Metric between Measures on Euclidean and Hilbert Spaces , 1990 .

[16]  J. Berman,et al.  Diffusion MR tractography as a tool for surgical planning. , 2009, Magnetic resonance imaging clinics of North America.

[17]  Ali R. Khan,et al.  The DTI Challenge: Toward Standardized Evaluation of Diffusion Tensor Imaging Tractography for Neurosurgery , 2015, Journal of neuroimaging : official journal of the American Society of Neuroimaging.

[18]  Vittorio Murino,et al.  From Covariance Matrices to Covariance Operators: Data Representation from Finite to Infinite-Dimensional Settings , 2016 .

[19]  Hervé Delingette,et al.  GPSSI: Gaussian Process for Sampling Segmentations of Images , 2015, MICCAI.

[20]  M. Knott,et al.  On the optimal mapping of distributions , 1984 .

[21]  Vittorio Murino,et al.  Log-Hilbert-Schmidt metric between positive definite operators on Hilbert spaces , 2014, NIPS.

[22]  I. Olkin,et al.  The distance between two random vectors with given dispersion matrices , 1982 .

[23]  Timothy Edward John Behrens,et al.  Effects of image reconstruction on fiber orientation mapping from multichannel diffusion MRI: Reducing the noise floor using SENSE , 2013, Magnetic resonance in medicine.

[24]  Mehrtash Tafazzoli Harandi,et al.  Bregman Divergences for Infinite Dimensional Covariance Matrices , 2014, 2014 IEEE Conference on Computer Vision and Pattern Recognition.

[25]  D. Dowson,et al.  The Fréchet distance between multivariate normal distributions , 1982 .

[26]  Essa Yacoub,et al.  The WU-Minn Human Connectome Project: An overview , 2013, NeuroImage.

[27]  Piercesare Secchi,et al.  Distances and inference for covariance operators , 2014 .

[28]  S Roberts,et al.  Gaussian processes for time-series modelling , 2013, Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences.

[29]  William Arveson,et al.  A Short Course on Spectral Theory , 2001 .

[30]  David Duvenaud,et al.  Probabilistic ODE Solvers with Runge-Kutta Means , 2014, NIPS.

[31]  Pierre Vandergheynst,et al.  Geodesic Convolutional Neural Networks on Riemannian Manifolds , 2015, 2015 IEEE International Conference on Computer Vision Workshop (ICCVW).

[32]  Rachid Deriche,et al.  Unsupervised white matter fiber clustering and tract probability map generation: Applications of a Gaussian process framework for white matter fibers , 2010, NeuroImage.

[33]  R. Tatusko Cooperation in climate research : an evaluation of the activities conducted under the US-USSR agreement for environmental protection since 1974 , 1990 .

[34]  Yoav Zemel,et al.  Procrustes Metrics on Covariance Operators and Optimal Transportation of Gaussian Processes , 2018, Sankhya A.

[35]  Mark Jenkinson,et al.  The minimal preprocessing pipelines for the Human Connectome Project , 2013, NeuroImage.

[36]  L. Ambrosio,et al.  A User’s Guide to Optimal Transport , 2013 .

[37]  Søren Hauberg,et al.  Probabilistic Shortest Path Tractography in DTI Using Gaussian Process ODE Solvers , 2014, MICCAI.