A simple multiple-fold correlation-based multi-view multi-label learning

Changming Zhu, Jiyun Zhao, Shizhe Hu, Yilin Dong, Lei Cao, Fan Zhou, Yuhu Shi, Lai Wei, Rigui Zhou

Abstract: Correlations among different features and labels are ubiquitous in the present multi-view multi-label data sets and they are always described with within-view, cross-view, and consensus-view representations. While how to discover and measure these correlations effectively so as to enhance performances of a learning machine is an open problem, this problem cannot be solved by existing traditional learning machines. In this article, different from the current classical multi-view, multi-label, multi-view multi-label learning machines, we focus on the simultaneously measurement of multiple-fold correlations including within-view ones, cross-view ones, and consensus-view ones. Then, a simple multiple-fold correlation-based multi-view multi-label learning (MC-MVML) is developed. Extensive experiments on 36 classical data sets validate the superiority of MC-MVML in terms of classification performance, training time, convergence, statistical analysis, influence of parameters, etc., and the development of multi-view multi-label learning theory is expected to be promoted.