迁移学习是机器学习和数据挖掘中的重要研究领域。在迁移学习中，知识从相关任务向目标任务迁移，并对目标任务的学习起到促进作用。本项目拟设计基于参数的支持向量机迁移学习算法，对迁移学习中的若干问题进行研究。首先，针对样本包含不确定信息的迁移学习问题，本项目通过最近邻思想把不确定信息边界引入学习，从而降低分类器对不确定信息的敏感性。其次，针对多示例分类中的知识迁移问题，构建基于相似度度量的多示例迁移学习算法，实现相关任务知识到目标任务的有效迁移。再次，针对文本数据流中同一时间窗的数据分布可能发生变化的问题，本项目自动侦测数据分布变化情况，并构造迁移学习分类器把变化前后的样本信息引入学习，从而提高分类器的预测能力。最后，针对异常点侦测的迁移学习问题，通过设计分类器之间的耦合参数，实现相关任务的分类器知识到目标任务分类器的迁移。本项目旨在通过以上问题的解决为迁移学习算法的研究提供思路。

Transfer learning is an important area in the machine learning and data mining communities. In transfer learning, the knowledge transferred from the related tasks to the target task can help the learning of the target task. This project aims at designing parameters-based support vector machine transfer learning algorithms to address the several research problems in transfer learning. First, to resolve the problem that the data set may contain uncertain data, the project proposes the usage of nearest neighbor idea to introduce the bound of data uncertainty into the learning of transfer learning so as to reduce the effect of uncertain data on the classifier. Second, the project puts forward similarity-based transfer learning approach to address the problem of multi-instance transfer learning problem such that knowledge can be transferred from source tasks to the target task. Third, to address the problem that in the textual data streams environment, the data distribution of the stream data in the same time window may change, this project automatically determines the change spot of data distribution in the time window, and then constructs a transfer learning classifier to introduce the data distribution between the change spot into learning. Last, to address the problem of classifier knowledge transfer in anomaly detection, the project determines the common parameters between the related tasks and target task, and then builds a transfer learning classifier to transfer knowledge to target task. This project aims at delivering research ideas to the area of transfer learning.