Semi-supervised Learning Based on Improved Co-training by Committee

摘要

As a popular machine learning technique, semi-supervised learning can make full use of a large pool of unlabeled samples in addition to a small number of labeled ones to improve the performance of supervised learning. In co-training by committee, a semi-supervised learning algorithm, the class probability values predicted by committee may repeat, which brings a negative influence on the improvement of the classification performance. We propose a method to deal with this problem, which assign different class probability estimations for different unlabeled samples. Naive Bayes is employed to help estimate the class probabilities of unlabeled samples. To prove that our method can reduce the introduction of noise, a data editing technique is employed to make a comparison with our method. Experimental results verify the effectiveness of our method and the data editing technique, and also indicate that our method is generally better than the data editing technique.