论文部分内容阅读
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.
For the high-dimensional data sets, there often exists the redundancy and dimensionality disaster, and directly construct the cover model which is difficult to fully reflect the data distribution information.A model based on sparse dimensionality reduction is proposed.Firstly, the homotopy algorithm is used to solve the sparse Which means that the optimization problem in the l_1, sparse constraints automatically obtain the number of reasonable neighbors and build the map, and then through the LPP (Locality Preserving Projections) to hold the local projection, and then to achieve high-dimensional space quickly and effectively reduce the dimension, and finally in low-dimensional space through An approximate convex hull coverage is constructed to implement a class classification.The experimental results on the UCI database, the MNIST handwritten database and the MIT-CBCL face recognition database confirm the effectiveness of the proposed method. Compared with the existing classifications, The model has a higher classification accuracy.