自适应标记关联与实例关联诱导的缺失多视图弱标记学习

Adaptive label correlation and instance correlation guided incomplete multiview weak label learning

针对多视图多标记学习中视图不完整和标记不完整问题,提出一种自适应标记关联与实例关联诱导的缺失多视图弱标记学习模型。模型假设样本各视图特征基于一个共享表示,通过不同映射得到。首先通过嵌入指示矩阵进行矩阵分解,充分利用已有的不完整多视图弱标记数据,然后引入图论中学习标准拉普拉斯矩阵的技术来刻画标记关联关系、实例关联关系,从而在模型里嵌入流形正则化思想,使学到的潜在共享表示以及分类器更加合理,最后在4个多视图多标记数据集上实验。实验结果表明,所提方法能够有效解决不完整多视图弱标记学习问题。

Focusing on the problem of incomplete view and label in multiview multilabel learning,the paper proposes a model called adaptive label correlation and instance correlation guided incomplete multiview weak label learning.The model assumes each view of the instance is obtained from a common representation through different mappings.Firstly,matrix factorization is used by embedding the indicator matrix to make full use of the existing incomplete multiview weak label data and then introduces the technology of learning the standard Laplacian matrix in graph theory to describe the label correlation and instance correlation.Therefore,embedding manifold regularization in the model to make the learned common representation and classifier more reasonable.Finally,four multiview multilabel datasets were conducted in a series of experiments.The experimental results show that the proposed model can solve the task of incomplete multiview weak label learning effectively.