面向多视角多标签数据的支持张量机分类算法研究

项目名称： 面向多视角多标签数据的支持张量机分类算法研究

项目编号： No.61472089

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： 郝志峰

作者单位： 广东工业大学

项目金额： 86万元

中文摘要： 面向多视角多标签数据的分类问题是目前模式识别、数据挖掘、图像处理、信息检索和生物信息学等领域广泛关注和研究的重点和难点课题。本项目拟利用统计学习理论、张量学习理论、多标签学习方法、核方法以及支持向量机算法模型，从理论、技术、应用三个层面，系统地开展面向多视角多标签数据的分类算法研究。提出基于特征选择与多标签分类器设计融合的支持张量机模型；给出相应的分类器性能评估准则；设计适用于大规模数据场景的在线增量学习算法；针对张量空间构造新的核函数，并探索结构自适应的多核学习方法；研究多视角下的标签修正算法，提高标签的质量。同时根据具体的应用需求进一步扩展和增强这些算法模型，并在国际公开的测试集数据上进行评测。这些研究将探索和发展新的多视角多标签分类算法的设计与分析的研究，并为其在相关领域的应用奠定理论基础和技术支撑，为研究以支持向量机为代表的的机器学习算法开拓新的理论视角。

中文关键词： 多视觉多标签数据；支持向量机；张量；特征选择与分类器融合；核方法

英文摘要： With the expansion and deepening of the application of classification analysis, the problem of multi-view multi-label classification has become a critical and challenging topic of current research in many scientific disciplines, such as pattern recognition, data mining, image processing, information retrieval and bioinformatics. In this proposal, we will focus on the systematic development of theory, methods and their applications for solving multi-view multi-label classification problems based on statistical learning theory, tensor analysis techniques, multi-label learning methods, kernel methods and support vector machine (SVM) algorithms. More specifically, (1) propose a series of support tensor machine models by interconnecting feature selection with the multi-label classifier design, (2) establish performance evaluation criteria to evaluate the performance of obtained classifier models, (3) design the online incremental learning algorithms suitable for large-scale analysis, (4) put forward new kernels in the tensor product space and explore multiple kernel learning scheme associated with adaptive structural changes, (5) study label correcting algorithms under multiple veiws to improve the qualityof labels, and evaluate our approaches on public data, and further adjust and enhance our models based on the specific applications. This proposal could benefit to explore and develop the study of design and analysis of algorithms for multi-view multi-label classification,while leading to some novel techniques and fundamental theoretical basis for their applications in various fields, and providing a new theoretical perspective for SVM-based machine learning algorithms.

英文关键词： Multi-view Multi-label Data;Support Vector Machines;Tensor;Feature Selection and Classifier Ensemble;Kernel Method