aiTPR: Attribute Interaction-Tensor Product Representation for Image Caption

Region visual features enhance the generative capability of machines based on features. However, they lack proper interaction-based attentional perceptions and end up with biased or uncorrelated sentences pieces misinformation. In this work, we propose Attribute Interaction-Tensor Product Representation (aiTPR), which is a convenient way gathering more information through orthogonal combination learning interactions as physical entities (tensors) improving captions. Compared to previous works, where add undefined feature spaces, TPR helps maintain sanity in combinations, orthogonality define familiar spaces. We have introduced new concept layer that defines objects their can play crucial role determining different descriptions. The interaction portions contributed heavily better caption quality out-performed various works domain MSCOCO dataset. For first time, notion combining regional image abstracted likelihood embedding for captioning.