A local domain adaptation feature extraction method

Abstract

In this paper, we propose a novel measure: Local Patches Based Maximum Mean Discrepancy (LPMMD). Based on the above measure, we also propose a novel feature extraction method: A Local Domain Adaptation Feature Extraction Method (LDAFE), which not only fulfills the transfer learning task, but also has a certain local learning capability. The LDAFE can complete traditional feature extraction as well as domain adaptation learning in two domains whose distributions are different but relative, thus indicating its better robustness and adaptation. Tests show the above-proposed advantages of the LPMMD criterion and the LDAFE method.