200 likes | 313 Views
Nonparametric Divergence Estimators for Independent Subspace Analysis. Barnabás Póczos (Carnegie Mellon University, USA) Zoltán Szabó (E ö tv ö s Lor á nd University, Hungary) Jeff Schneider (Carnegie Mellon University, USA). EUSIPCO‐2011 Barcelona, Spain Sept 2, 2011. Outline.
E N D
Nonparametric Divergence Estimators for Independent Subspace Analysis Barnabás Póczos (Carnegie Mellon University, USA) Zoltán Szabó (Eötvös Loránd University, Hungary) Jeff Schneider (Carnegie Mellon University, USA) EUSIPCO‐2011 Barcelona, Spain Sept 2, 2011
Outline • Goal:divergence estimation • Definitions, basic properties, motivation • The estimator • Theoretical results • Consistency • Experimental results • Mutual information estimation • Independent subspace analysis • Low-dimensional embedding of distributions
Measuring divergences Manchester United 07/08 Owen Hargreaves Rio Ferdinand Cristiano Ronaldo KL Tsallis Rényi www.juhokim.com/projects.php
Density: nuisance parameterDensity estimation: difficult How should we estimate them? • Naïve plug-in approach using density estimation • density estimators • histogram • kernel density estimation • k-nearest neighbors [D. Loftsgaarden & C. Quesenberry. 1965.] • How can we estimate them directly?
kNN density estimation How good is this estimation? [D. Loftsgaarden and C. Quesenberry. 1965.] [N. Leonenko et. al. 2008]
Asymptotically unbiased The estimator We need to prove: Agner Krarup Erlang 1-, and -1 moments of the “normalized k-NN distances” Normalized k-NN distances converge to the Erlang distribution 7
Asymptotically unbiased If we could move the limit inside the expectation… All we need is
Solutions: Asymptotically uniformly integrability… A little problem… Increases the paper length by another 20 pages…
Results for divergence estimation 2D Normal 10
Results for MI estimation rotated uniform distribution
Independent Subspace Analysis Independent subspaces 6 by 6 mixing matrix Observation X=AS Estimate A and S observing samples from X only Goal: 12
Independent Subspace Analysis Objective: 13
Low dimensional embeddig of digits Noisy USPS datasets
Be careful, some mistakes are easy to make… We want: Helly–Bray theorem [Annals of Statistics]
Fatou lemma: Erlang Fatou lemma: [Journal of Nonparametric Statistics, Problems Information Transmission, IEEE Trans. on Information Theory] Some mistakes … We want: Enough:
Takeaways If you need to estimate divergences, then use me! • Consistent divergence estimator • Direct: no need to estimate densities • Simple: it needs only kNN based statistics • Can be used for mutual information estimation, independent subspace analysis, low-dimensional embedding Thanks for your attention!