基于_2,1范数的超图正则化非负矩阵分解

Hypergraph Regular Nonnegative Matrix Factorization Based on _2,1 Norms

图学习的方法只考虑了样本间的关系,忽略了样本间的高维结构,基于欧氏距离的非负矩阵模型易受到噪声和异常值的影响。为解决以上问题,给出一种基于L2,1范数的超图正则化非负矩阵分解(L2,1HNMFL)方法,并给出了算法的更新规则。通过在COIL20和ORL经典数据集上与其他算法的比较验证了该算法的有效性。

The graph learning method only considers the relationship between samples, but ignores the high-dimensional structure between samples. The non-negative matrix model based on Euclidean distance is easily affected by noise and outliers. To solve the above problems, a hypergraph regu-larization nonnegative matrix factorization (L2,1HNMFL) method based on L2,1 norm is presented, and the updating rules of the algorithm are given. The effectiveness of this algorithm is verified by comparing it with other algorithms on classical data sets COIL20 and ORL.