计算机科学 ›› 2012, Vol. 39 ›› Issue (11): 212-215.

• 人工智能 • 上一篇    下一篇

融合稀疏保持的成对约束投影

齐鸣鸣,向阳   

  1. (同济大学计算机科学与技术系 上海201804);(绍兴文理学院元培学院 绍兴312000)
  • 出版日期:2018-11-16 发布日期:2018-11-16

Pairwise Constraint Projections Inosculating Sparsity Preserving

  • Online:2018-11-16 Published:2018-11-16

摘要: 提出一种融合稀疏保持的成对约束投影(Pairwise Constraint Projections inosculating Sparsity Preserving, SPPCP)。该算法在成对约束指导的降维过程中,通过平衡参数引入稀疏保持投影(Sparsity Preserving Projections, SPP),在保持成对约束特征的同时,也继承了稀疏保持所蕴含的几何结构保持和近部保持特性。在UCI数据集和 AR人脸库上的实验表明,该算法有效地融合了稀疏保持投影的优点,与典型的成对约束的半监督降维算法相比,提 高了基于最短欧氏距离的分类算法的精度和稳定性。

关键词: 成对约束,稀疏保持,半监督降维

Abstract: A kind of algorithm called Pairwise Constraint Projections inosculating Sparsity Preserving (SPPCP) was proposed,which introduces Sparsity Preserving Projections(SPP) with trade-off parameter in the process of pairwise constraint-guided dimensional reduction,preserving pairwise constraint feature and heriting the special character of geo- metrical structure preserving and neighbor preserving from sparsity preserving. Experiments operated on UCI databases and AR face database show that the algorithm inosculates merits of SPP effectively. Compared with the typical semi su- pervised dimensional reduction algorithms based on pairwise constraint, the algorithm can improve the accuracy and sta- bility of classified algorithms based on the shortest Euclidean distance.

Key words: Pairwise constraints, Sparsity preserving, Semi-supervised dimensional reduction

No related articles found!
Viewed
Full text


Abstract

Cited

  Shared   
  Discussed   
No Suggested Reading articles found!