不充分视图半监督学习的理论分析研究

In real-world applications, multiple views come along with a large amount of unlabeled data. However, each of the views may not be sufficient due to feature corruption or various noise. In this research project, we focus on the theoretical study on semi-supervised learning with insufficient views. We try to provide the theoretical analysis for co-training with insufficient views, try to provide the theoretical analysis for co-training with insufficient views under Tsybakov condition, try to complement the theoretical analysis for co-regularization and study the difference between co-training and co-regularization, and develop new semi-supervised learning algorithms dealing with insufficient views based on the theoretical results. It is expected to publish 4-6 papers on important international journals and conferences and native top journals, apply 1-2 patents, and supervise 2 graduate students.

在实际应用中，多视图往往伴随大量未标记数据同时出现，然而由于属性的退化和各种噪声的影响，每个视图可能都是不充分的，本项目关注于不充分视图半监督学习的理论研究，针对半监督学习中最重要的风范之一协同训练，给出不充分视图协同训练的理论分析；针对目前最接近真实问题的理论设置Tsybakov 条件，给出Tsybakov 条件下不充分视图协同训练的理论分析；针对协同正则化，完善不充分视图协同正则化的理论分析并研究其与协同训练的本质区别；依据理论分析得到的结果，设计并实现不充分视图半监督学习算法。本项目可望在国际期刊、会议和国内一级学报上发表论文4-6篇，申请专利1-2项，培养研究生2名。

在实际应用中，多视图往往伴随大量未标记数据同时出现，然而由于属性的退化和噪声的影响，每个视图可能都是不充分的，本项目关注于不充分视图半监督学习的理论研究，给出不充分视图协同训练的理论分析；给出Tsybakov 条件下不充分视图协同训练的理论分析；针对协同正则化，完善不充分视图协同正则化的理论分析并研究其与协同训练的本质区别；依据理论分析得到的结果，设计并实现了不充分视图半监督学习算法。本项目在国际期刊、会议和国内核心期刊共发表论文7篇，培养研究生3名。