项目名称: 对称锥互补问题的算法研究及其在压缩感知中的应用
项目编号: No.11426168
项目类型: 专项基金项目
立项/批准年度: 2015
项目学科: 数理科学和化学
项目作者: 李远敏
作者单位: 西安电子科技大学
项目金额: 3万元
中文摘要: 利用若当代数技术研究对称锥互补问题是国内外优化界的研究热点。随着应用的不断深入,目前存在的求解对称锥互补问题的迭代算法已不能满足实际需要,无法进行实时求解,而神经网络方法是进行实时求解的非常有效的手段。然而,国内外尚未见到求解对称锥互补问题的神经网络方法的相关文献报道。本课题创造性地建立求解对称锥互补问题的神经网络方法和半光滑牛顿算法,并利用该方法建立压缩感知信号重建新算法。项目解决的关键问题包括:建立若当代数上互补函数的次微分理论,发展求解对称锥互补问题的半光滑牛顿算法;借助对称锥互补函数,建立求解对称锥互补问题的神经网络模型;选取并优化压缩感知信号重建模型,利用神经网络方法和半光滑牛顿算法恢复重建原始信号。本项目的研究成果将进一步推动对称锥互补问题的发展,为对称锥互补问题的实际应用提供理论依据;为对称锥互补问题与国际研究热点压缩感知找到新的切入点,为压缩感知信号重建提供新的研究思路。
中文关键词: 若当代数;神经网络;光滑算法;互补问题;信号重建
英文摘要: It is a hot research subject in optimization domain to study symmetric cone complementarity problems with help of Jordan algebraic technique. With the development of applications, the existing iterative algorithms can not meet the actual needs. For example, they can not obtain real time solution. Though neural network method is an effective means for real time, it has not yet been reported in literature home and aboard to solve symmetric cone complementarity problems. This topic will creatively establish neural network method and semi-smooth Newton method to solve symmetric cone complementarity problems, and then apply these methods to develop new signal reconstruction algorithms for compressed sensing. Key issues addressed by the project include the following three aspects. First, set up subdifferentiable theory of complementarity functions over Jordan algebras and design semismooth Newton algorithm. Then, establish neural network method to solve symmetric cone complementarity problems with symmetric cone complementarity functions. Last, select and optimize the compressed sensing signal reconstruction model, and then recover the original signal with neural network method and semismooth Newton method. The scientific significance of the research results will not only further promote the development of symmetric c
英文关键词: Jordan algebra;neural network;smoothing method;complementarity problem;signal reconstruction