基于辅助学习与富信息策略的Tri-training算法  

AR-Tri-training: Tri-training with assistant and rich information strategy

作　　者：崔龙杰[1,2] 王红丽 崔荣一[1] 

机构地区：[1]延边大学计算机科学与技术系智能信息处理研究室,吉林延吉133002 [2]延边大学财务处,吉林延吉133002 [3]山东优加利信息技术有限公司,济南250101

出　　处：《计算机应用研究》2014年第9期2685-2687,共3页Application Research of Computers

基　　金：吉林省教育厅"十二五"科学技术研究项目(吉教科合字[2011]第8号)

摘　　要：针对Tri-training算法利用无标记样例时会引入噪声且限制无标记样例的利用率而导致分类性能下降的缺点,提出了AR-Tri-training(Tri-training with assistant and rich strategy)算法。提出辅助学习策略,结合富信息策略设计辅助学习器,并将辅助学习器应用在Tri-training训练以及说话声识别中。实验结果表明,辅助学习器在Tri-training训练的基础上不仅降低每次迭代可能产生的误标记样例数,而且能够充分地利用无标记样例以及在验证集上的错分样例信息。从实验结果可以得出,该算法能够弥补Tri-training算法的缺点,进一步提高测试率。Tri-training introduced noise samples and restricted the utilization of unlabeled samples for the use of unlabeled samples, so the classification performance of the classifier decreased. This paper proposed a new Tri-training style algorithm named AR-Tri-training （Tri-training with assistant and rich strategy）. Firstly, it presented the assistant learning strategy. Sec- ondly, it designed the supporting learner by combining the assistant learning strategy with rich information strategy. Finally, the training process of Tri-training and voice recognition applied the supporting learner. The experimental results show that the supporting learner not only reduces the number of mislabeled samples produced in the iterations based on Tri-training, but also makes full use of the unlabeled samples and the misclassified samples of validation set. It can be drawn from the experimental results that AR-Tri-training algorithm can compensate for the shortcomings of Tri-training algorithm, further improve the testing rate.

关 键 词：半监督学习 富信息策略 辅助学习策略 Tri—training 说话声识别 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] TP301.6[自动化与计算机技术—计算机科学与技术]