Random Moments for Sketched Mixture Learning

Nicolas Keriven (1) , Rémi Gribonval (1) , Gilles Blanchard (2) , Yann Traonmilin (1)

1 PANAMA - Parcimonie et Nouveaux Algorithmes pour le Signal et la Modélisation Audio
2 Institut für Mathematik [Potsdam]

We present a method to solve large-scale mixture learning tasks from a sketch of the data, formed by random generalized empirical moments. We give empirical and theoretical results on k-means and Gaussian Mixture Model estimation problems.

Domaines: Machine Learning [stat.ML]; Apprentissage [cs.LG]

Nicolas Keriven, Rémi Gribonval, Gilles Blanchard, Yann Traonmilin. Random Moments for Sketched Mixture Learning. SPARS2017 - Signal Processing with Adaptive Sparse Structured Representations workshop, Jun 2017, Lisbon, Portugal. ⟨hal-01494045⟩