Gene functional classification from heterogeneous data
RECOMB '01 Proceedings of the fifth annual international conference on Computational biology
Multiple kernel learning, conic duality, and the SMO algorithm
ICML '04 Proceedings of the twenty-first international conference on Machine learning
A statistical framework for genomic data fusion
Bioinformatics
Large Scale Multiple Kernel Learning
The Journal of Machine Learning Research
Localized multiple kernel learning
Proceedings of the 25th international conference on Machine learning
Multiple Kernel Learning Algorithms
The Journal of Machine Learning Research
Hi-index | 0.00 |
Kernel classifiers have demonstrated their high performance for many classification problems. For the proper selection of kernel functions, multiple kernel learning (MKL) has been researched. Furthermore, the localized MKL (LMKL) enables to set the weights for the kernel functions at each point. However, the training of the weight functions for kernel functions is a complex nonlinear problem and a classifier can be trained separately after the weights are fixed. The iteration of the two processes are often necessary. In this paper we propose a new framework for MKL/LMKL. In the framework, not kernel functions but mappings to the feature space are combined with weights. We also propose a new learning scheme to train simultaneously weights for kernel functions and a classifier. We realize a classifier by our framework with the Gaussian kernel function and the support vector machine. Finally, we show its advantages by experimental results.