mPLUG: Effective and Efficient Vision-Language Learning by Cross-modal Skip-connections. (EMNLP 2022)
