Semantic-Layer: A High-Level Semantic Ontology For Data-Driven Approaches

Recent years have seen a breakthrough in object detection in computer vision thanks to the development of deep learning algorithms. Although neural approaches appear to perform on par with or even better than human judgment for several benchmarks, issues remain for identifying random concepts in random fields. In this paper, we propose and construct a high-level semantic ontology called the “semantic-layer”, which can be used as an inference engine as well as in training for data-driven models, such as, computer vision, image retrieval, and NLP (Natural language processing) systems. In addition, the proposed layer is also destined to query generation systems, which aim to modify or enhance the users' queries before being transmitted to the server of the research engines.