Traditional machine learning methods, such as Support Vector Machines (SVMs), usually assume that training and test data share the same distributions. Due to the inherent dynamic data nature, it is often observed that (1) ...