Many machine classification problems resemble human decision making tasks, which are multimodal in nature. Humans have the capability to combine various types of sensory data and associate them with natural language entities. Complex decision tasks such as person identification often heavily depend on such synergy or fusion of different modalities. For that reason, much effort on machine classification methods goes into exploiting the underlying relationships among modalities and constructing an effective fusion algorithm. This is a fundamental step in the advancement of artificial intelligence, because the scope of learning algorithms is not limited to one type of sensory data.