跨领域整体模式分类理论研究及应用

项目名称： 跨领域整体模式分类理论研究及应用

项目编号： No.61473236

项目类型： 面上项目

立项/批准年度： 2015

项目学科： 其他

项目作者： 黄开竹

作者单位： 西交利物浦大学

项目金额： 75万元

中文摘要： 本项目针对多个异构领域的非独立同分布数据，开展跨领域整体模式分类理论方法与应用研究。跨领域整体模式识别方法是相对较新的课题，旨在充分利用领域间异构数据的互补性和相关性，并打破领域内数据独立同分布的假设，不独立地对每一样本进行分类，而是将多领域协同学习，并对一组具有相同特性的领域内数据进行整体处理和同时分类，从而有效利用领域与领域之间、数据与数据之间的联系，提高识别性能。跨领域整体模式分类具有很强的理论意义和学术影响，有助于推动模式识别和机器学习领域的发展；同时又具有重要的应用价值，可被广泛应用于各种实际系统之中，为国家经济建设和国防安全等服务。

中文关键词： 模式分类；跨领域；整体识别；非i.i.d.数据

英文摘要： The objective of this project is to develop cross-domain and collective pattern classification theory and its applications for heterogeneous non i.i.d data. The proposed methodology is a new research theme, and attemps to take advantages of the complementary and relevant information between different domains. Moreover, breaking the i.i.d. assumption in traditional pattern recognition methods, our project does not classify each data sample separately and isolatedly. In contrast, this research attempts to classify a group of within-domain data samples (sharing the same charasteristics) simultaneously in a collective way. The cross-domain and collective pattern classification method is capable of better utilizing both the between-domain and within-domain information from data, and hence can provide potentials in improving the learning accuracy substantially. Our proposed research is of great significance in theory and will make big impact in pattern recognition and machine learning. It is also of great value in practice and is expected to be applied extensively in many real systems so as to better serve the economy and national defense of our country.

英文关键词： Pattern Classification;Cross Domain;Collective Recognition;Non i.i.d.Data