Gao Quanxue, Liu Jingjing, Cui Kai, Zhang Hailin, Wang Xiaogang
State Key Laboratory of Integrated Services Networks, Xidian University, Xi'an, China.
State Key Laboratory of Integrated Services Networks, Xidian University, Xi'an, China.
Neural Netw. 2014 Jun;54:49-56. doi: 10.1016/j.neunet.2014.02.009. Epub 2014 Mar 4.
Locality Sensitive Discriminant Analysis (LSDA) is one of the prevalent discriminant approaches based on manifold learning for dimensionality reduction. However, LSDA ignores the intra-class variation that characterizes the diversity of data, resulting in unstableness of the intra-class geometrical structure representation and not good enough performance of the algorithm. In this paper, a novel approach is proposed, namely stable locality sensitive discriminant analysis (SLSDA), for dimensionality reduction. SLSDA constructs an adjacency graph to model the diversity of data and then integrates it in the objective function of LSDA. Experimental results in five databases show the effectiveness of the proposed approach.
