南京理工大学学报(自然科学版)
南京理工大學學報(自然科學版)
남경리공대학학보(자연과학판)
JOURNAL OF NANJING UNIVERSITY OF SCIENCE AND TECHNOLOGY
2014年
4期
518-525
,共8页
特征选择%半监督学习%成对约束%结构和约束保持%特征排序%空间结构%先验知识
特徵選擇%半鑑督學習%成對約束%結構和約束保持%特徵排序%空間結構%先驗知識
특정선택%반감독학습%성대약속%결구화약속보지%특정배서%공간결구%선험지식
feature selection%semi-supervised learning%pairwise constrains%structure and constraints preserving%feature ranking%geometrical structure%supervision information
针对现有特征选择算法较少同时考虑样本的空间结构和先验知识的不足,提出一种基于结构和约束保持的半监督特征选择方法。该方法采用成对约束作为先验知识,同时考虑局部和非局结构,定义了一种新的特征评价准则---结构和约束保持分值。利用大量的无标记样本来学习样本空间结构,利用少量的成对约束信息来学习类内和类间边缘,所选择的特征子集能较好地保持空间结构信息和类属信息。在多个数据集上的实验结果表明,和现有的几种特征排序选择算法相比,所提方法有较好表现。
針對現有特徵選擇算法較少同時攷慮樣本的空間結構和先驗知識的不足,提齣一種基于結構和約束保持的半鑑督特徵選擇方法。該方法採用成對約束作為先驗知識,同時攷慮跼部和非跼結構,定義瞭一種新的特徵評價準則---結構和約束保持分值。利用大量的無標記樣本來學習樣本空間結構,利用少量的成對約束信息來學習類內和類間邊緣,所選擇的特徵子集能較好地保持空間結構信息和類屬信息。在多箇數據集上的實驗結果錶明,和現有的幾種特徵排序選擇算法相比,所提方法有較好錶現。
침대현유특정선택산법교소동시고필양본적공간결구화선험지식적불족,제출일충기우결구화약속보지적반감독특정선택방법。해방법채용성대약속작위선험지식,동시고필국부화비국결구,정의료일충신적특정평개준칙---결구화약속보지분치。이용대량적무표기양본래학습양본공간결구,이용소량적성대약속신식래학습류내화류간변연,소선택적특정자집능교호지보지공간결구신식화류속신식。재다개수거집상적실험결과표명,화현유적궤충특정배서선택산법상비,소제방법유교호표현。
To overcome the deficiency of most existing feature selection methods which fairly respect both the geometrical structure and the supervision information, a novel approach called semi-supervised feature selection based on structure and constraints preserving is proposed. In this method,both the pairwise constraints and the local and nonlocal structure are taken into account,and a new feature selection criterion,i. e. structure and constraints preserving( SCP) score is defined. The SCP score exploites abundant unlabeled data points to learn the geometrical structure of the data space,and uses a few pairwise constraints to discover the margins of different classes. Those features that can preserve the geometrical structure and pairwise constraints information are selected. Experimental results from several datasets show that the proposed method achieves better performance than the feature ranking selection methods.