从希尔伯特-施密特独立性中学习的多标签半监督学习方法

Multi-label semi-supervised learning method learnt from Hilbert-Schmidt independence criterion

基于希尔伯特-施密特独立性提出了一种新的半监督学习方法,称为最大化依赖性多标签半监督学习方法(dependence maximization multi-label semi-supervised learning method,DMMS)。该方法将样本已有标签作为约束,以最大化特征集和标签集的关联性为目标,通过求解一个线性系统为无标签数据打上标签,具有实现简单,无参(nonparameter)的特点。多个真实多标签数据库的实验表明,DMMS与最好的多标签学习方法,包括多标签近邻(multi-label k-nearest neighbor,MLKNN)和图半监督学习方法具有类似的识别效果。

Hilbert-Sehmidt independence criterion （HSIC） can be used to measure the correlation degree of feature set and label set of samples. On the basis of HSIC, this paper presents a new semi-supervised learning method called dependence maximization multi-label semi-supervised learning method （DMMS）. By setting the existing labels as constraint and dependence of features and labels as optimization objective, the method solves a linear system to get the labels for unlabeled samples, possessing the features of simple implementation and no parameter estimation. Experiments on some real multi-label datasets show that the proposed method is as good as the state-of-the-art multi-label learning methods in recognition tasks, including multi-label k-nearest neigh- bor （MLKNN） and graph based semi-supervised learning method.