Representative Multi-Label Bayesian Approach for image classification

Recently, multi-label learning approaches are gaining more and more attention due to its useful applications in the area of data mining and bioinformatics. Though there exist a lot of multi-label learning approaches, few of them consider how to deal with the dataset with noisy attributes. In this paper, we will present Representative Multi-Label Bayesian Approach (RMLBA) to process the dataset with noisy attributes. RMLBA incorporates the affinity propagation (AP) approach and the Bayesian approach into the multi-label learning framework. Instead of considering all the attributes, RMLBA only focuses on a small subset of representative attributes which is detected by the AP. The experiments on image classification illustrate the RMLBA works well for the multi-label classification problems.