Nowadays, huge amount of multimedia data are being constantly generated in various forms from various places around the world. With ever increasing complexity and variability of multimedia data, traditional rule-based approaches where humans have to discover the domain knowledge and encode it into a set of programming rules are too costly and incompetent for analyzing the contents, and gaining the intelligence of this glut of multimedia data. The challenges in data complexity and variability have led to revolutions in machine learning techniques. In the past decade, we have seen many new developments in machine learning theories and algorithms, such as boosting, regressions, Support Vector Machines, graphical models, etc