A survey of transfer learning

This survey paper addresses transfer learning, a methodology that departs from the traditional machine learning assumption that training and testing data come from the same domain with identical feature space and distribution characteristics. Because that assumption fails in scenarios where training data is expensive or difficult to collect, transfer learning aims to create high-performance learners trained with more easily obtained data drawn from different domains. The paper formally defines transfer learning and presents information on current solutions.

The survey reviews applications of transfer learning and provides practical resources, including information on software downloads for various transfer learning solutions, plus a discussion of possible future research directions. A key point emphasized is that the surveyed transfer learning solutions are independent of data size and can therefore be applied to big data environments, broadening their practical relevance across real-world machine learning and data mining applications.