1. Transfer Learning Classification
(1) Instance-based transfer learning method:
- Hypothesis: Some data in the source domain and the target domain will share many common features
- Method: Perform instance reweighting on the source domain, filter out the data with high similarity with the target domain data, and then carry out training and learning
(2) Feature-based transfer learning
When the source domain and the target domain contain some common cross-features, we can transform the features of the source domain and the target domain into the same space through feature transformation, so that the source domain data in this space and the target domain The domain data has the same distribution as the data distribution, and then traditional machine learning.
(3) Model-based migration learning
In model-based migration, the source domain and the target domain share model parameters, that is, the model previously trained with a large amount of data in the source domain is applied to the target domain for prediction.
- Features: The same part of the model can be directly migrated
- no training data required
(4) Relation-based migration learning
When two domains are similar, they will share a similar relationship, and apply the logical network relationship learned in the source domain to the target domain for migration, such as biological viruses Migration of Propagation Laws to Computer Virus Propagation Laws. There is relatively little research work in this part. The typical method is the method of mapping.
总结迁移的方式：
- data
- feature
- Model
- train of thought