A Comprehensive Survey on Transfer Learning

This survey provides a comprehensive review of transfer learning, which aims to improve the performance of target learners on target domains by transferring knowledge contained in different but related source domains, thereby reducing dependence on large amounts of target-domain data. Unlike prior surveys that introduce approaches in a relatively isolated way, this article attempts to connect and systematize existing studies and to summarize and interpret the mechanisms and strategies of transfer learning. It reviews more than 40 representative approaches, with particular emphasis on homogeneous transfer learning, organized from the perspectives of data and model.

Beyond categorizing methods and briefly introducing applications, the survey empirically compares over 20 representative transfer learning models on three datasets—Amazon Reviews, Reuters-21578, and Office-31. The experimental results demonstrate the importance of selecting appropriate transfer learning models for different applications in practice, giving readers both a systematic map of the field and practical evidence to guide model choice.