Self-training method using cluster information of unlabeled samples
Xin-Shun Xu · Jisuanji yingyong yanjiu · 2010
In order to use the structure information,this paper proposed a new method proposed in which used a data editing method and a clustering method to remove the most likely mislabeled samples and selected confident ones from self-labeled samples. Experimental results on UCI machine learning repository show that the performance and convergence are better and faster than the contrastive algorithm. It shows that introduction of data clustering to select sample is beneficial.