Bottom-Up and Top-Down Attention for Image Captioning and Visual Question Answering | IEEE Conference Publication | IEEE Xplore