熊凯^1，， 韩军伟^2，*

（  1、 西北工业大学自动化学院,西安市 710072 ；       2、 西北工业大学自动化学院,西安市 710072；  ）

摘要： 特征选择在模式识别、机器学习以及其它相关领域起着重要的作用。特征选择用于数据预处理阶段,通过选择一个具有区分性和高信息量的特征子集实现高效的降维。近年来,基于正则化的特征选择引起研究者极大的兴趣,并且已经有很多算法被提出来了。这些方法的共同思想是给特征选择向量或选择矩阵选择一个合适的范数,比如$ell_1$范数、$ell_1/ell_infty$范数和$ell_{2,1}$范数来实现稀疏性,选择向量中的非零元或者选择矩阵中的非零行就对应了被选中的特征。本文旨在对近年来提出的代表性工作进行综述,提供关于这些方法之间联系与区别的观点,并对当前面临的挑战和未来可能的工作方向进行讨论。
关键词： 模式识别；特征选择；正则化

Kai Xiong^1,， Junwei Han^2,*

（  1、 School of Automation, Northwestern Polytechnical University, Xi'an 710072 ；       2、 School of Automation, Northwestern Polytechnical University, Xi'an 710072；  ）

Abstract： Feature selection plays a significant role in pattern recognition, machine learning and other related fields. It can be used in data preprocessing step to achieve efficient dimensionality reduction by selecting a subset of discriminant and informative features. In recent years, regularization based feature selection has attracted great interest and many algorithms have been proposed. The common idea of these methods is to choose a proper norm such as $ell_1$-norm, $ell_1/ell_infty$-norm and $ell_{2,1}$-norm for the selection vector or selection matrix to achieve sparseness, and the non-zero entries or rows correspond to the selected features. This paper aims to review the recently proposed representative works, provides insight into the differences and connections between these methods, and discusses the challenges and possible directions of future work.

Tag：

点此返回栏目查看更多>>>参考论文