ABSTRACT
Department of Computer Science and Engineering, Hong Kong University of Science and Technology, Hong Kong
Yu Zhang
Department of Computer Science and Engineering, Hong Kong University of Science and Technology, Hong Kong
Qiang Yang
Department of Computer Science and Engineering, Hong Kong University of Science and Technology, Hong Kong
In this chapter we discuss transfer learning and multi-task learning problems, and relate them to cost-sensitive learning. In many machine learning problems, the learning problem in one or more domains of interest, known as target domains, may be very difficult to solve due to a lack of high-quality labeled training data, but we may have some related knowledge from one or more different but similar domains. In such cases, we may find some common knowledge between these domains to help improve the learning performance in some chosen target domains, or improve the performance of learning in all related domains. Learning under these circumstances is called transfer learning or multi-task learning (see a survey by Pan and Yang [42]).
