核函数优化选择的关键技术研究

项目名称： 核函数优化选择的关键技术研究

项目编号： No.61202265

项目类型： 青年科学基金项目

立项/批准年度： 2013

项目学科： 计算机科学学科

项目作者： 汪廷华

作者单位： 赣南师范学院

项目金额： 22万元

中文摘要： 以支持向量机(Support Vector Machine,SVM)为主要代表的核方法(Kernel Methods)目前在模式识别、函数估计等领域获得了广泛的应用。核函数是影响核方法性能的关键因素，从核函数度量(简称核度量)这个切入点来研究核函数优化选择的关键技术及其应用，以达到进一步提升核方法性能和拓展核方法应用范围的目的。主要研究内容包括：1)从特征空间数据点的分布特性出发，设计有效的核度量标准的技术；2)基于核度量的高效的分阶段多核学习(Multiple Kernel Learning,MKL)算法；3)多核学习在自然语言处理词义消歧(Word Sense Disambiguation,WSD)中的应用。本项目的研究不仅对支持向量机具有重要的意义，而且对其它有监督的核学习模型的性能改进也具有重要的应用价值，同时对于如何根据特定的应用领域选择使用核函数问题也具有一定的借鉴意义。

中文关键词： 核函数选择；核度量；支持向量机；多核学习；词义消歧

英文摘要： Kernel methods have been successfully apllied for a wide range of different data analysis problems, such as pattern recognition and function estimation. A typical example of kernel methods is support vector machine (SVM). It is well known that kernel function is a crucial factor of achieving good performance for kernel methods. In this work, taking the issue of kernel function evaluation as cut-in piont, we will investigate some key technologies for kernel optimization and selection, in order to further improve the performance and expand the application range of kernel methods. To be specific, in this work we will: 1) study the approaches to design effective kernel evaluation criterion,which is based on the data distribution in the feature space; 2) present efficient stage-by-stage multiple kernel learning algorithm based on kernel evaluation; and 3) study the application of multiple kernel learning to word sense disambiguation in natural language processing. This work is valuable for SVM and other supervised kernel-based learning models, as well as the issue that how to choose appropriate kernel function according to the specific application domain.

英文关键词： kernel selection；kerne evaluation；support vector machine (SVM)；multiple kernel learning (MKL)；word sense disambiguation (WSD)