The performance improvement algorithm of co-training by committee
Ruiyan Wang, Li Li · 2016
Semi-supervised learning is a key research subject in the field of machine learning. Co-training by Committee is an iterative semi-supervised learning algorithm. During the iteration of this algorithm, the previous committee is used for predicting unlabeled examples. However, the classification accuracy limitation of single committee will bring adverse effect on training of committees. Therefore, under the intuition that multiple committees own higher classification accuracy, an improved algorithm that utilizes all the previous committees to predict examples at each iteration is proposed. The algorithm is designed to provide labeled data set with more properly labeled examples, consequently, promote the performance of generated committees. This improvement algorithm can use multiple committees effectively. The classification accuracy of Co-training by Committee increases more than ten percentage points in average.