Part-based transfer learning

Abstract

Transfer learning, serving as one of the most popular theory in machine learning, has attracted a lot of attention recently. In this paper, we propose a new learning strategy called part-based transfer learning (pbTL), which is a process of parameter transfer. Dissimilar to many traditional works, we consider how to transfer the information from one task to another in the form of parts. We regard all the complex tasks as a collection of constituent parts and every task can be divided into several parts respectively. It means transfer learning between two complex tasks can be accomplished by sub-transfer learning tasks between their parts. Through developing it in this hierarchical fasion, we can reach a better outcome. Experiments on synthetic data with support vector machines (SVMs) validate the effectiveness of the proposed learning framework. © 2011 Springer-Verlag.