Transfer learning through domain adaptation

Abstract

It is interesting and helpful to use the labeled data of some tasks to improve the classification performance of another task. This paper focuses on this issue and proposes an algorithm named SSDT (Synthetic Source Data Transfer). As the number of the training data influences the classification performance greatly, we create some synthetic training data using the source data and combine them with the target data to train a classifier. The classifier is applied to the target data, and experimental results show that SSDT improves the performance obviously. © 2011 Springer-Verlag.