Learning the Structure of Similarity