Stochastic Low-Rank Kernel Learning for Regression

Pierre Machart (1, 2) , Thomas Peel (3, 1) , Liva Ralaivola (3) , Sandrine Anthoine (1) , Hervé Glotin (2)

1 LATP - Laboratoire d'Analyse, Topologie, Probabilités
2 LSIS - Laboratoire des Sciences de l'Information et des Systèmes
3 QARMA - éQuipe AppRentissage et MultimediA [Marseille]

We present a novel approach to learn a kernel-based regression function. It is based on the useof conical combinations of data-based parameterized kernels and on a new stochastic convex optimization procedure of which we establish convergence guarantees. The overall learning procedure has the nice properties that a) the learned conical combination is automatically designed to perform the regression task at hand and b) the updates implicated by the optimization procedure are quite inexpensive. In order to shed light on the appositeness of our learning strategy, we present empirical results from experiments conducted on various benchmark datasets.

Mots clés: Stochastic Optimization; Kernel Learning; Low-Rank Approximation

Domaines: Apprentissage [cs.LG]

Pierre Machart, Thomas Peel, Liva Ralaivola, Sandrine Anthoine, Hervé Glotin. Stochastic Low-Rank Kernel Learning for Regression. International Conference on Machine Learning (ICML'11), Jun 2011, Bellevue (Washington), United States. pp.969--976, ISBN : 978-1-4503-0619-5. ⟨hal-00657837⟩