Research Of Kernel Methods For Support Vector Machine And Multiple Kernel Clustering Algorithm

Since Vapnik proposed the Support Vector Machine (SVM) based on Statistical Learning Theory and kernel trick in 1995,kernel method based machine learning algorithm has been developed rapidly.It becomes one of the hot points in academic research now and has been widely used in image processing,biology information technology, intrision detection and text classification,etc.So it is of great significance for both the development and improvement of kernel theory and its expansion of application.Kernel Function, as one of the most important ways of non-linear mapping, is the essence of Support Vector Machines with such wide application. An independent discipline called Kernel Methods has been formed especially for kernel functions. The purpose of this thesis is to research the properties and construction of multiple kernel functions. Research of kernel functions does not only improve the usage of Support Vector Machines, but also gives support to Artificial Intelligence and Machine Learning themselves.This paper has mainly discussed the following problems:(1) Multiple kernels is proposed due to learning problems involve multiple and heterogeneous data sources. Choosing different parameters of different kernel functions or kernel function according to different properties to improve learning ability and generalization of kernel,and prove the legitimacy of the new kernel.(2) A fuzzy clustering algorithm based on multiple kernels is proposed due to learning problems involve multiple and heterogeneous data sources.The method overcomes limitations of the algorithm based on single kernel and improves the learning and generalization performance. By using multiple kernel functions, the original data can be mapped into a high-dimensional feature space so that the important characteristics of data are shown clearer. The clustering experiments show that multiple kernels fuzzy clustering algorithm can behave more effectively than that of single kernel.(3) Support Vector Clustering belongs to unsupervised learnig machine models and its result depends entirely on the selected kernel function. Gaussian kernel is commonly used in support vector clustering. However, the limitations of the Gaussian kernel width determines the generalization ability of kernel machine. Therefore propose support vector clustering algorithm based on Weighted Gaussian Kernel with Multiple Widths, which can achieve better clustering results.