项目名称: 面向高维信息的非线性维数约简问题研究
项目编号: No.61303091
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: 高小方
作者单位: 山西大学
项目金额: 22万元
中文摘要: 本项目针对具有大规模、多批次、多流形特征的高维信息数据,系统研究面向非线性维数约简问题的流形学习方法,及其在图像数据和生物信息数据中的应用。具体内容包括:(1)构建面向多流形数据的切空间构造方法,并基于采样密度和流形曲率计算动态邻域;(2)提出一种基于切空间扩展的多流形识别并分解的方法,并扩展多流形数据的应用研究;(3)给出一种具有增量学习能力的流形学习方法。最终,本项目形成一个面向高维信息的非线性维数约简系统,并应用于图像识别、图像检索、生物信息比对等领域。本项目的研究将进一步丰富流形学习方法的理论和算法研究,而且将为机器学习、图像识别、生物信息学等相关领域的应用研究提供技术支持。
中文关键词: 流形学习;非线性维数约简;增量学习;多流形;
英文摘要: This project systematically studies manifold learning methods for non-linear dimensionality reduction of high-dimensional information including large-scale, arriving in batches, multi-manifolds datasets. Firstly, an efficient method to construct tangent spaces for multi-manifolds is proposed, and a dynamical neighborhood graph model based on local sampling density and manifold curves is computed. Secondly,a method to decompose and recognize multi-manifolds by propagating the tangent spaces is devised. Then we propose a novel incremental manifold learning method. Finally, a nonlinear dimensionality reduction system for high-dimensional information is developed, which can be applicated to image recognition, information retrieval and biomedicine. The implentation of this project would further enrich the researches of manifold learning, and generalize its application in the fields of machine learning, image recognition and bioinformatics.
英文关键词: Manifold Learning;Nonlinear Dimensionality Reduction;Incremental Learning;Multi-Manifolds;