基于图正则平滑低秩表示的基因表达谱特征选择

FEATURE SELECTION OF GENE EXPRESSION DATA BASED ON SMOOTHED LOW-RANK WITH GRAPH REGULARIZATION

针对传统低秩表示不能准确描述数据结构的问题,提出一种图正则平滑低秩表示的特征选择算法。在构造目标函数时,利用对数行列式函数代替核范数来平滑估计秩函数,引入流形正则项;利用非精确交替方向法进行求解,并且采用后处理方式构造数据的图结构。该算法能够准确地描述数据全局子空间结构和局部线性结构。在基因表达谱数据集上进行聚类实验,同其他特征选择算法相比较,实验结果证明了该算法的有效性。

An improved feature selection algorithm of graph regularization smoothed low rank representation is proposed to solve the problem of inaccuracy of describing data structure using traditional low-rank representation.When constructing the objective function,the logarithm determinant function is used to do smoothing estimation instead of Kernel function,and manifold regularized item is added.The objective function is solved by ADM and the graph construction is restructured by a post-processing method.This algorithm can describe global subspace structure and partial linear structure accurately.After clustering experiment in gene expression profile and comparing with other feature selection algorithm,the results verify the effectiveness of the proposed approach.