Joint Learning of Scene Graph Generation and Reasoning for Visual Question Answering Mid-term report