项目名称: 快速卷积型张量分解理论研究及在fMRI处理中的应用
项目编号: No.61305060
项目类型: 青年科学基金项目
立项/批准年度: 2014
项目学科: 自动化技术、计算机技术
项目作者: 吴强
作者单位: 山东大学
项目金额: 27万元
中文摘要: 张量作为一种有力的高阶数据分析工具,能够有效地对多因素数据进行表征。研究张量分解模型以及快速分解算法对大规模复杂数据表征和处理具有重要的理论价值,在磁共振影像分析、脑电信号处理、模式识别等领域具有重要的应用。本项目针对功能磁共振影像处理中的成分延迟问题,研究具有延迟特性的卷积型张量分解理论,拟提出新型的张量分解模型,引入新的优化代价函数,研发快速学习算法,并解决相关的理论问题,如计算效率、稳定性和收敛性等。针对大规模功能磁共振影像特征选取问题,利用张量低秩分解和局部自适应采样策略,提出新型的快速分解算法。研究功能磁共振成像的时延模型,利用快速卷积型张量分解算法进行多因素关联分析,提出针对阿尔茨海默病的功能磁共振影像体素选择和特征提取框架,为发病机理研究和辅助诊断提供依据。
中文关键词: 卷积张量分解;稀疏表征;磁共振影像分析;特征提取与选择;大规模数据处理
英文摘要: As a powerful tool for higher order data analysis, tensor can represent the multiple factors data efficiently. Tensor factorization model and fast factorization algorithms have significant theoretic value for complicated large-scale data representation and processing. And they have important applications in many fields including MRI analysis, brain signal processing and pattern recognition. In this project, we make a study of component delay problem in fMRI processing and plan to investigate the convolutive tensor factorization theory considering delay. New tensor factorization structure, cost function and fast learning algorithms will be proposed and some theoretical analysis of algorithms such as computational efficiency, stability, convergence property will be presented. In order to solve the feature extraction problem in large-scale fMRI processing, we plan to develop new fast factorization algorithms using low rank decomposition and adaptive local sampling criteria. The time delay model for fMRI data will be investigated and we employ fast convolutive tensor factorization algorithms to perform multiple factors association analysis. Through voxel selection and feature extraction on fMRI data of Alzheimer's disease, supplementary evidence will be provided for pathogenesis and aided clinical diagnosis.
英文关键词: Convolutive Tensor Decomposition;Sparse Representation;MRI Analysis;Feature Extraction and Selection;Largescale Data Analysis