球面学习理论研究

项目名称： 球面学习理论研究

项目编号： No.60873206

项目类型： 面上项目

立项/批准年度： 2009

项目学科： 轻工业、手工业

项目作者： 曹飞龙

作者单位： 中国计量学院

项目金额： 30万元

中文摘要： 在许多实际应用中, 数据往往在球面上采集，人们试图寻找“#26426;器结构”#30340;泛函模型来“#22788;理”#12289;“#20998;析”#12289;“#29983;产”#25152;收集到的数据,从中获取所需的“#30693;识”#12290;而基于数据的机器学习是当前信息处理中的重要技术并已成为信息科学的研究热点之一。本项目围绕这一目标，对球面上学习机器的构造、学习的理论与算法等核心问题展开深入、系统的研究。我们汲取并发展球调和分析、函数逼近等某些重要的理论与方法，创造性地建立球面神经网络、支持向量机等学习机器的一般构造方法，多角度、着重研究了球面学习速率估计的理论与方法，构建球面学习的理论及算法。本项研究的完成不但为诸多实际问题的解决提供理论与方法，而且将发展、丰富学习理论，乃至球调和分析、球面函数逼近的理论,进一步推动交叉学科的发展。本项研究发表论文47篇，其中被SCI检索24篇，EI检索14篇，组织国际学术会议 “nternational Symposium on Extreme Learning Machines”一次。课题组共参加国际学术会议6人次，邀请国内专家作学术报告8人次，招收硕士研究生14名，毕业硕士研究生9名。

中文关键词： 机器学习；数据分类；逼近；球面

英文摘要： In many applied problems such as seismology, earth measure and cerebral medical simulation, data is often collected on a spherical surface. So people try to look for a functional model with “achine architecture”to handle analysis and manufacture those data collected to get the information needed. Based on machine learning of data is the most important technology in information handling and one of the hot topic in informatics, this project researched on the following core problems such as construction of learning machine, learning theory and algorithm and data classification on a spherical surface thoroughly and systematically. It absorbed and developed some important theory and algorithm of harmonic analysis, function approximation and statistics, established a general architecture method for the neural network and support vector machine on a spherical surface, and studied the theory and method of spherical surface learning rate estimation variously to construct the theory and algorithm of spherical surface learning. Hence the accomplishment of this project not only provided theory and method for applied problems, but also developed and enriched learning theory, which promoted the development of cross subjects. Finishing the item results in that forty-seven papers have been published, where the twenty-four papers were cited by SCI, and fourteen papers were cited EI. In the process of the item to be carried out, a conference tiled by International Symposium on Extreme Learning Machines was organized, six person-times taken part in the interpolation conference, and eight experts were invited to give lectures. Meanwhile, fourteen graduate students ware recruited, and nine graduate students have gotten master degree.

英文关键词： Machine learning; Classification of data; Approximation; Sphere