Wide consensus aggregation in the Wasserstein space. Application to location-scatter families

We introduce a general theory for a consensus-based combination of estimations of probability measures. Potential applications include parallelized or distributed sampling schemes as well as variations on aggregation from resampling techniques like boosting or bagging. Taking into account the possibility of very discrepant estimations, instead of a full consensus we consider a “wide consensus” procedure. The approach is based on the consideration of trimmed barycenters in the Wasserstein space of probability measures. We provide general existence and consistency results as well as suitable properties of these robustified Fréchet means. In order to get quick applicability, we also include characterizations of barycenters of probabilities that belong to (non necessarily elliptical) location and scatter families. For these families we provide an iterative algorithm for the effective computation of trimmed barycenters, based on a consistent algorithm for computing barycenters, guarantying applicability in a wide setting of statistical problems. AMS Subject Classification: Primary: 60B05, 62F35, Secondary 62H12.

[1]  P. Rousseeuw.  Least Median of Squares Regression , 1984 .

[2]  P. Rousseeuw.  Multivariate estimation with high breakdown point , 1985 .

[3]  J. A. Cuesta,et al.  The strong law of large numbers for k-means and best possible nets of Banach valued random variables , 1988 .

[4]  J. A. Cuesta,et al.  Notes on the Wasserstein Metric in Hilbert Spaces , 1989 .

[5]  S. Rachev,et al.  A characterization of random variables with minimum L 2 -distance , 1990 .

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

[7]  Y. Brenier.  Polar Factorization and Monotone Rearrangement of Vector-Valued Functions , 1991 .

[8]  A. Gordaliza.  Best approximations to random variables based on trimming procedures , 1991 .

[9]  J. A. Cuesta-Albertos,et al.  Optimal coupling of multivariate distributions and stochastic processes , 1993 .

[10]  M. Knott,et al.  On a generalization of cyclic monotonicity and distances among random vectors , 1994 .

[11]  David M. Rocke,et al.  Computable Robust Estimation of Multivariate Location and Shape in High Dimension Using Compound Estimators , 1994 .

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

[13]  Christophe Croux,et al.  An easy way to increase the finite-sample efficiency of the resampled minimum volume ellipsoid estimator , 1997 .

[14]  C. Czado,et al.  Nonparametric validation of similar distributions and assessment of goodness of fit , 1998 .

[15]  Katrien van Driessen,et al.  A Fast Algorithm for the Minimum Covariance Determinant Estimator , 1999, Technometrics.

[16]  C. Matrán,et al.  A central limit theorem for multivariate generalized trimmed $k$-means , 1999 .

[17]  P. Bühlmann,et al.  Analyzing Bagging , 2001 .

[18]  L. Rüschendorf,et al.  On the n-Coupling Problem , 2002 .

[19]  Dudley,et al.  Real Analysis and Probability: Measurability: Borel Isomorphism and Analytic Sets , 2002 .

[20]  Carlos Matrán Bea,et al.  Shape of a Distribution Through the L2-Wasserstein Distance , 2002 .

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

[22]  P. Bühlmann.  Bagging, subagging and bragging for improving some prediction algorithms , 2003 .

[23]  G. Burton.  TOPICS IN OPTIMAL TRANSPORTATION (Graduate Studies in Mathematics 58) By CÉDRIC VILLANI: 370 pp., US$59.00, ISBN 0-8218-3312-X (American Mathematical Society, Providence, RI, 2003) , 2004 .

[24]  Leo Breiman,et al.  Bagging Predictors , 1996, Machine Learning.

[25]  Nicholas Ayache,et al.  Geometric Means in a Novel Vector Space Structure on Symmetric Positive-Definite Matrices , 2007, SIAM J. Matrix Anal. Appl..

[26]  J. A. Cuesta-Albertos,et al.  Trimming and likelihood: Robust location and dispersion estimation in the elliptical model , 2008, 0811.0503.

[27]  C. Villani.  Optimal Transport: Old and New , 2008 .

[28]  J. A. Cuesta-Albertos,et al.  Uniqueness and approximate computation of optimal incomplete transportation plans , 2011 .

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

[30]  Thibaut Le Gouic,et al.  Distribution's template estimate with Wasserstein metrics , 2011, 1111.5927.

[31]  Brendan Pass.  Optimal transportation with infinitely many marginals , 2012, 1206.5515.

[32]  J. A. Cuesta-Albertos,et al.  Similarity of samples and trimming , 2012, 1205.1950.

[33]  Luis Angel García-Escudero,et al.  tclust: An R Package for a Trimming Approach to Cluster Analysis , 2012 .

[34]  Jérémie Bigot,et al.  Consistent estimation of a population barycenter in the Wasserstein space , 2013 .

[35]  V. Chernozhukov,et al.  Monge-Kantorovich Depth, Quantiles, Ranks and Signs , 2014, 1412.8434.

[36]  Adam M. Oberman,et al.  NUMERICAL METHODS FOR MATCHING FOR TEAMS AND WASSERSTEIN BARYCENTERS , 2014, 1411.3602.

[37]  Arnaud Doucet,et al.  Fast Computation of Wasserstein Barycenters , 2013, ICML.

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

[39]  Jean-Michel Loubes,et al.  Barycenter in Wasserstein Spaces: Existence and Consistency , 2015, GSI.

[40]  Thibaut Le Gouic,et al.  Existence and consistency of Wasserstein barycenters , 2015, Probability Theory and Related Fields.

[41]  Gabriel Peyré,et al.  Iterative Bregman Projections for Regularized Transportation Problems , 2014, SIAM J. Sci. Comput..

[42]  Carlos Matrán,et al.  Robust k-barycenters in Wasserstein Space and Wide Consensus Clustering , 2016 .

[43]  Axel Munk,et al.  Limit laws of the empirical Wasserstein distance: Gaussian distributions , 2015, J. Multivar. Anal..

[44]  Peter Bühlmann,et al.  Magging: Maximin Aggregation for Inhomogeneous Large-Scale Data , 2014, Proceedings of the IEEE.

[45]  Juan Antonio Cuesta-Albertos,et al.  Robust clustering tools based on optimal transportation , 2016, Statistics and Computing.