Learning geometric combinations of Gaussian kernels with alternating Quasi-Newton algorithm

David Picard (1) , Nicolas Thome (2) , Matthieu Cord (2) , Alain Rakotomamonjy (3)

1 MIDI - Multimedia Indexation and Data Integration
2 MALIRE - Machine Learning and Information Retrieval
3 LITIS - Laboratoire d'Informatique, de Traitement de l'Information et des Systèmes

We propose a novel algorithm for learning a geometric com- bination of Gaussian kernel jointly with a SVM classifier. This problem is the product counterpart of MKL, with restriction to Gaussian kernels. Our algorithm finds a local solution by alternating a Quasi-Newton gradi- ent descent over the kernels and a classical SVM solver over the instances. We show promising results on well known data sets which suggest the soundness of the approach.

Domains: Machine Learning [stat.ML]

David Picard, Nicolas Thome, Matthieu Cord, Alain Rakotomamonjy. Learning geometric combinations of Gaussian kernels with alternating Quasi-Newton algorithm. 20th European Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning, Apr 2012, Bruges, Belgium. pp.79-84. ⟨hal-00705374⟩