Novel Inductive and Transductive Transfer Learning Approaches Based on Support Vector Learning

Abstract

In this chapter, two novel transfer learning approaches based on support vector learning are involved. For inductive transfer learning, the knowledge-leverage-based TSK fuzzy system (KL-TSK-FS) is proposed, which demonstrates the good privacy-protection abilities and strong adaptability for the situations where the data are only partially available from the target domain while some useful knowledge of the source domains is available. For transductive transfer learning, domain adaptation kernelized support vector machine (DAKSVM) and its two extensions are proposed, which can reduce the distribution gap between different domains in an RKHS as much as possible by integrating the large margin learner with the proposed generalized projected maximum distribution distance (GPMDD) metric.