基于稀疏分解的局部全局一致性学习算法

Learning with local and global consistency based on sparse representation

提出一种基于稀疏分解的l0构图法,通过稀疏分解系数矩阵得到图中邻接矩阵和边的权重。将l0构图法应用到局部全局一致性学习(Learning with Local and Global Consistency,LLGC)算法中,并通过K-均值聚类优化稀疏分解所需字典,以降低计算复杂度。在8个UCI数据集上的实验表明,与经典LLGC算法相比,新算法能在消耗时间不增加的情况下提高分类精度,提升算法性能。

Abstract. A graph construction method by the sparse representation is proposed in this paper. This method derives graph adjacency structure and the graph weights from coefficient matrix sim- ultaneously. The graph is introduced into Learning with Local and Global Consistency （LLGC） algorithm. The dictionary is optimized by K-means algorithm. Computing complexity can then be reduced and the performance of LLGC can be improved. Experimental results on UCI datasets in- dicate that the proposed algorithm can improve the precision of classification and maintain con- sumed time invariant. Therefore the performance of LLGC is improved.