12-in-1: Multi-Task Vision and Language Representation Learning - 42Papers