Recurrent Neural Symbolic Execution Module

AbstractThe predominant approach of visual question answering (VQA) relies on encoding the imageand question with a ”black box” neural encoder and decoding a single token into answers suchas ”yes” or …...