Transfer LearningIt is a method of using existing knowledge to learn new knowledge, where the original knowledge is called the source domain, and the new knowledge to be learned is the target domain. The purpose of transfer learning is to extract knowledge and experience from the source tasks and apply them to the target domain.
Transfer learning classification
Based on feature space:
- Homogeneous transfer learning: The feature space of the source domain and the target domain is the same, XS = XT;
- Heterogeneous transfer learning: The feature spaces of the source domain and the target domain are different, XS ≠ XT.
Based on the migration scenario:
- Inductive transfer learning: the learning tasks in the source domain and the target domain are different;
- Transductive transfer learning: The learning tasks in the source domain and the target domain are the same;
- Unsupervised transfer learning: Both the source domain and the target domain have no labels.
Basic methods of transfer learning
- Sample migration: Find data in the source domain that is similar to the target domain and adjust the data weights so that the new data matches the target domain data;
- Model transfer: Assuming that the source domain and the target domain share model parameters, the model trained in the source domain is applied to the target domain for prediction;
- Relationship migration: Assuming that the two domains are similar, the logical network relationship in the source domain is applied to the target domain for migration.
References
【1】https://zhuanlan.zhihu.com/p/27368456
【2】https://github.com/jindongwang/transferlearning/blob/master/doc//Introduction to Transfer Learning.md
【3】http://www.xtecher.com/Xfeature/view?aid=7383