Comix: joint estimation and lightspeed comparison of mixture models

Conférence Chapitre d'actes Comix: joint estimation and lightspeed comparison of mixture models eng 2016 author Schwander Olivier 838159 author Marchand-Maillet Stéphane 0000-0002-4875-6101 518563 author Nielsen Frank 2449-2453 10.1109/ICASSP.2016.7472117 978-1-4799-9988-0 http://ieeexplore.ieee.org/document/7472117/ The Kullback-Leibler divergence is a widespread dissimilarity measure between probability density functions, based on the Shannon entropy. Unfortunately, there is no analytic formula available to compute this divergence between mixture models, imposing the use of costly approximation algorithms. In order to reduce the computational burden when a lot of divergence evaluations are needed, we introduce a sub-class of the mixture models where the component parameters are shared between a set of mixtures and the only degree-of-freedom is the vector of weights of each mixture. This sharing allows to design extremely fast versions of existing dissimilarity measures between mixtures. We demonstrate the effectiveness of our approach by evaluating the quality of the ordering produced by our method on a real dataset. Mixture model Density estimation Information geometry Kullback-Leibler divergence Exponential family Dewey 025.063 Piscataway (N.J.) IEEE 13 Centre universitaire d'informatique 0D3 32 Département d'informatique 18 false 895 Computer Vision and Multimedia Laboratory 675 Viper group IEEE International Conference on Acoustics, Speech and Signal Processing, ICASSP 2016 20-25 March 2016 Shanghai (China) d96a752e-6817-401e-abae-1ce1edf7a0f5 Proceedings chapter (Published version) Restricted 2016Schwander_comix.pdf 480655 application/pdf restrictedAccess Centres et instituts / Centre universitaire d'informatique Faculté des sciences / Département d'informatique

Title	Published in	Access level	OA Policy	Year	Views	Downloads

Comix: joint estimation and lightspeed comparison of mixture models	IEEE International Conference on Acoustics, Speech and Signal Processing, ICASSP 2016			2016	357	0

Archive ouverte UNIGE

Schwander, Olivier