Transfer Learning Methods and Their Application in Radar Signal Processing

Abstract

Transfer learning method uses the similarity of data, tasks, models in different domains and applies the knowledge from one domain to another domain. Traditional machine learning requires that the test and training data sets obey the same data distribution. In contrast, transfer learning can reuse the training samples of similar data distribution in different fields and improve new fields' learning and training effect. Currently, there are many studies on the theory of transfer learning, especially in the field of domain adaptability. This paper summarizes and analyzes the research progress of data distribution adaptation methods, feature selection methods, subspace learning methods, and deep transfer learning methods. Then, the application of transfer learning in the radar field is further introduced. Finally, the future research direction of transfer learning is discussed.