一种面向多源领域的实例迁移学习

在迁移学习最大的特点就是利用相关领域的知识来帮助完成目标领域中的学习任务，它能够有效地在相似的领域或任务之间进行信息的共享和迁移，使传统的从零开始的学习变成可积累的学习，具有成本低、效率高等优点.针对源领域数据和目标领域数据分布类似的情况，提出一种基于多源动态TrAdaBoost的实例迁移学习方法.该方法考虑多个源领域知识，使得目标任务的学习可以充分利用所有源领域信息，每次训练候选分类器时，所有源领域样本都参与学习，可以获得有利于目标任务学习的有用信息，从而避免负迁移的产生.理论分析验证了所提算法较单源迁移的优势，以及加入动态因子改善了源权重收敛导致的权重熵由源样本转移到目标样本的问题.实验结果验证了此算法在提高识别率方面的优势.

Instance-based Transfer Learning for Multi-source Domains

The most remarkable characteristic of transfer learning is that it can employ the knowledge in relative domains to help perform the learning tasks in the domain of the target. With the use of different fields of knowledge for target task learning, transfer learning can transfer and share the information between similar domains or tasks, making the traditional learning from scratch an addable one, which implies that the learning efficiency is higher and the cost is lower. For the specific situation that the shared knowledge in the domains of the source and the target are sample data with similar distribution, an instance transfer learning method based on multi-sources dynamic TrAdaBoost is put forward. Integrated with the knowledge in multiple source domains, this method makes the target task learning the one that is able to make good use of the information of all source domains. Whenever candidate classifiers are trained, all the samples in all source domains are involved in learning, and the information conducive to target task learning can be obtained, so that negative transfer can be avoided. The theoretical analysis suggests that the given algorithm is better than the single source transfer. By means of adding the dynamic factor, this algorithm improves the defect that weight entropy drifts from source to target instances. The experimental results support that the given algorithm has the advantage of improving the recognition rate.