稀疏降维的近似凸壳覆盖一类分类器构造

A One-class Classifier Construction Based Approximate Convex Hull Covering Model Using Dimensionality Reduction By Sparse Representation

针对高维数据集常常存在冗余和维数灾难,在其上直接构造覆盖模型难以充分反映数据分布信息的问题,提出一种基于稀疏降维近似凸壳覆盖模型.首先采用同伦算法求解稀疏表示中l_1优化问题,通过稀疏约束自动获取合理近邻数并构建图,再通过LPP(Locality Preserving Projections)来进行局部保持投影,进而实现对高维空间快速有效地降维,最后在低维空间通过构造近似凸壳覆盖实现一类分类.在UCI数据库,MNIST手写体数据库和MIT-CBCL人脸识别数据库上的实验结果证实了方法的有效性,与现有的一类分类算法相比,提出的覆盖模型具有更高的分类正确率.

Considering redundant and curse of dimensionality in high-dimensional data, a covering model constructed from these data can not reflect their distributing information. To solve this problem, an approximate convex hull covering model based dimensionality reduction by sparse representation is proposed. Firstly the homotopy algorithm is used to solve e1 norm problem, neighbors are automatically captured based sparse constraint then neighborhood graph is constructed. Next, LPP is applied in order to fast and efficient dimensionality reduction. And finally, an approximate convex hull covering model is constructed in low-dimensional space and realized one-class classification. Experimental results show that the proposed covering method has better correct rate for classification by comparing with results of other one-class classification method on the UCI, MNIST and MIT-CBCL face data sets.