一种能够实现双层稀疏性的Group Logarithm模型  

The Group Logarithm Model with Bi-Level Sparsity

作　　者：崔立鹏 郭红霞 沈洁[1] CUI Lipeng;GUO Hongxia;SHEN Jie(School of Electronic Information and Automation,Tianjin Light Industry Vocational Technical College,Tianjin 300350,China)

机构地区：[1]天津轻工职业技术学院电子信息与自动化学院,天津300350

出　　处：《苏州市职业大学学报》2021年第2期40-45,共6页Journal of Suzhou Vocational University

基　　金：国家职业教育智能控制技术专业教学资源库建设项目(2018-12)。

摘　　要：稀疏化模型研究是人工智能领域研究的热点之一。对于该问题Lasso模型忽略了变量之间的组结构,只能实现分散的变量选择;Group Lasso也只能实现变量组选择。本研究提出Group Logarithm模型,该模型能同时进行变量组选择和组内的变量选择,并利用局部坐标下降算法求解Group Logarithm模型。最后,通过实验说明了该模型具有更优越的预测准确性和变量选择稀疏性。The research of sparse models is quite hot in the field of artificial intelligence in recent years.Lasso neglects the group structure in the variables and can only carry out distributed variable selection.Group Lasso can only achieve variable group selection.In order to solve this problem,the Group Logarithm model is proposed and the Group Logarithm performs variable group selection across the groups and variable selection within a group at the same time.Then,the local coordinate descent algorithm is used to solve the Group Logarithm model.In the end,the experiments show the advantages of the Group Logarithm in the respects of variable selection and prediction.

关 键 词：人工智能 稀疏性 变量选择 变量组选择 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]