| Kernel method is a machine learning algorithm that based on Statistical Learning Theory and kernel trick, and it can complete the nonlinear relationship between data with statistical analysis effectively. At present, it becomes one of the hot points in academic research. However, the conventional kernel methods are based on the single types of kernels, which often fail to describe the characteristics. Therefore, it is quite necessary to study the method based on multiple kernels. The learning algorithms for multi-kernel are expected to describe the diverse characteristics as detailed as possible by means of fusion of multiple kernel functions. This paper researched the two-stage multiple kernel learning, after studying the theory of multiple kernel learning, concluding the improved algorithm, two-stage multiple kernel learning with L2-regularization. Then several UCI databases are implemented to verify the performance of two-stage multiple kernel learning algorithm with L2-regularization. The main research contents of this paper were described as follows.In method study,this paper focuses on the study of multiple kernel learning algorithm, and also presents a comprehensive discussion about the major factors that influence the algorithm effect. This paper is developed by providing a detailed introduction of two-stage multiple kernel learning algorithms. To improve the performance of two-stage multiple kernel learning algorithm, this paper develops comprehensive study on weights of basic kernels. Then multiple kernel learning with L2-regularization is provided to maximize the optimization effect.In experimental verification, to provide a sufficient comparison between the algorithm leaned in this paper and classic ones, including LDA, KDA, TS-MKL-F and TS-MKL-B. This paper implements them for several UCI databases. A comprehensive analysis is presented after comparing results, which indicates that the proposed algorithm is 6.02%、3.43% and 1.30% than classic ones. |
PDF Full Text Request