Niveau: Supérieures
SimpleMKL Alain Rakotomamonjy LITIS EA 4108 Universite de Rouen 76800 Saint Etienne du Rouvray, France Francis Bach INRIA - Willow project Departement d'Informatique, Ecole Normale Superieure 45, Rue d'Ulm 75230 Paris, France Stephane Canu LITIS EA 4108 INSA de Rouen 76801 Saint Etienne du Rouvray, France Yves Grandvalet CNRS/IDIAP Research Institute, Centre du Parc, Av. des Pres-Beudin 20 1920 Martigny, Switzerland Abstract Multiple kernel learning aims at simultaneously learning a kernel and the associated pre- dictor in supervised learning settings. For the support vector machine, an efficient and general multiple kernel learning (MKL) algorithm, based on semi-infinite linear progam- ming, has been recently proposed. This approach has opened new perspectives since it makes the MKL approach tractable for large-scale problems, by iteratively using exist- ing support vector machine code. However, it turns out that this iterative algorithm needs numerous iterations for converging towards a reasonable solution. In this pa- per, we address the MKL problem through an adaptive 2-norm regularization formu- lation that encourages sparse kernel combinations. Apart from learning the combina- tion, we solve a standard SVM optimization problem, where the kernel is defined as a linear combination of multiple kernels.
- vector machine
- algorithm
- svm
- svm problems
- kernel learning
- norm regularization
- mkl problem
- like regression