In many problems of machine learning and computer vision, there exists side information, i.e., information contained in the training data and not available in the testing phase. This motivates the recent development of a new learning approach known as learning with side information that aims to incorporate side information for improved learning algorithms. In this work, we describe a new traini...