基于组稀疏的参数自适应学习UDP算法

Parameter adaptively learning UDP based on group spares

传统UDP算法的参数选择是一个经典问题,至今仍没有一种有效的方法从根本上解决这个问题。针对复杂的参数选择,提出一种基于组稀疏的参数自适应学习UDP算法(SUDP)。使用组稀疏来描述样本点的几何结构,自适应地构造样本点的近邻图,避免传统UDP算法中使用K-NN算法带来的弊端。由于稀疏表示带有天然判别信息的优势,SUDP算法比传统的UDP算法有着更强的判别能力。在6个广泛使用的人脸数据集上进行的实验,实验结果表明了SUDP算法的有效性和稳定性。

Parameter selection of traditional UDP algorithm is a classic problem,which is still unsolved in an effective way.To solve the problem,a parameter adaptively learning UDP based on group sparse was proposed,named as SUDP.The intrinsic geometry of data was described using group sparse representation,and the neighborhood graph was adaptively constructed.The weakness of traditional UDP of using K-NN to construct K-neighborhood graph disappeared in SUDP.Meanwhile,SUDP has better discriminant ability than UDP,because sparse representation has natural discriminant information.Experimental results on 6 widely used face databases show that SUDP is effective and stable.