A Survey on Transfer Learning

Abstract

To facilitate learning in a target domain, transfer learning borrows knowledge from a source domain. What and how to transfer are two main issues that need to be addressed in transferring learning. Different transfer learning algorithms result in different knowledge transferred between them for a couple of domains. To find the optimal transfer learning algorithm that maximizes learning efficiency in the target domain, scientists need to investigate all current computationally intractable transfer learning algorithms exhaustively. A sub-optimal algorithm is selected as a trade-off, which in an ad hoc way requires considerable expertise. In instructional psychology, meanwhile, it is commonly recognized that people enhance the transfer of teaching abilities to decide what to transfer. This paper discusses what is transfer learning, the different transfer learning techniques, future scope, and applications of it.