Representation space learned for multiple modalities or objects so semantically related items lie close together, used in retrieval and cross modal search.