基于■2,1范数和神经网络的非线性特征选择方法  

Nonlinear Solution for ■2,1 -Norm Based Feature Selection and Neural Network

作　　者：范馨予 徐雪远 邬霞[1] FAN Xinyu;XU Xueyuan;WU Xia(Beijing Normal University the School of Artificial Intelligence,Beijing 100875,China)

机构地区：[1]北京师范大学人工智能学院,北京100875

出　　处：《信号处理》2021年第9期1644-1652,共9页Journal of Signal Processing

摘　　要：在基于■2,1范数的特征选择方法中,■2,1范数可以使选择的特征具有组间稀疏性和组内稀疏性,同时还可以去除特征数据中的异常值。然而,大多数基于■2,1范数的特征选择算法常通过线性方程求解,无法探究特征之间的非线性关系。因此,本文提出了一种基于■2,1范数的非线性特征选择方法,将■2,1范数与神经网络相结合。一方面,该方法利用神经网络的非线性特性对■2,1范数进行求解。另一方面,该方法利用■2,1范数实现基于神经网络框架的特征选择。最后,本文将该方法与当前流行的特征选择方法在八个公开数据集进行了对比,实验结果验证了该方法具有一定的优越性。In joint ■2,1-norm based feature selection methods,predicted attributes shared parallel sparsity coefficients and outliers could be removed.However,previous optimizing ■2,1-norm solutions were usually based on linear solutions which could not discover the non-linear relationship between features.In this paper,we proposed a nonlinear solution for ■2,1-norm based feature selection using the framework of neural network.With combination of ■2,1-norm and neural network,on the one hand,we can make use of the nonlinearity of neural network to optimize ■2,1-norm based problem.On the other hand,the ■2,1-norm can help neural network to achieve feature selection.Finally,the proposed method is compared with famous feature selection methods on eight datasets.Experimental results empirically show the superiority of our method.

关 键 词：非线性特征选择 ■2 1范数 神经网络 后向传播 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]