基于核函数的低秩非线性属性选择算法

Low rank nonlinear feature selection algorithm via kernel method

高维的数据之间除了有线性关系之外,还有非线性的关系,为此提出一种基于核函数的低秩非线性属性选择算法。通过核函数把数据的每个属性都映射到高维空间,在高维空间中表示出属性之间的非线性关系;通过低秩约束排除噪声的干扰,利用高维空间中数据属性对应的稀疏正则化因子进行属性选择。核函数映射可以找出数据属性之间的非线性关系,低秩约束相当于对高维数据的全局结构进行子空间学习。该算法是一种嵌入了低秩结构的非线性属性选择算法。实验结果表明,该算法比其它的属性选择算法具有更好的效果。

Aiming at the linear relationship and the nonlinear relationship between high-dimensional data,a low-rank nonlinear feature selection algorithm via kernel function was proposed.Each feature of the data was mapped to high-dimensional space using a kernel function,and a nonlinear relationship between the features was expressed in high-dimensional space.The interfe-rence of noise was eliminated using the low rank constraint,and the features were selected based on the sparse regularization factor corresponding to the data feature in the high dimensional space.The kernel function mapping can find the non-linear relationship between data features.The low-rank constraint is equivalent to performing subspace learning on the global structure of high-dimensional data.The algorithm is a non-linear feature selection algorithm embedded with a low rank structure.The final expe-rimental results show that it has better effects than other feature selection algorithms.