Genetic Programming for Kernel-based Learning with Co-evolving Subsets Selection
Christian Gagné, Marc Schoenauer, Michele Sebag and Marco Tomassini
In: PPSN'06, 9-13 Sept 2006, Reykjavik.
Support Vector Machines (SVMs) are well-established Machine Learning (ML) algorithms. They rely on the fact that i) linear learning can b e formalized as a well-p osed optimization problem; ii) non-linear learning can b e brought into linear learning thanks to the kernel trick and the mapping of the initial search space onto an high dimensional feature space. The kernel is designed by the ML expert and it governs the eﬃciency of the SVMs approach. In this pap er, a new approach for the automatic design of kernels by Genetic Programming, called the Evolutionary Kernel Machine (EKM), is presented. EKM combines a well-founded ﬁtness function inspired from the margin criterion, and a co-evolution framework ensuring the computational scalability of the approach. Empirical validation on standard ML b enchmark demonstrates that EKM is comp etitive using state-of-the-art SVMs with tuned hyper-parameters.