Cornell Natural Language Visual Reasoning

AbstractTraining semantic parsers from question-answer pairs typically involves searching over an exponentially large space of logical forms, and an unguided search can easily be misled by spurious logical forms that …...