The author proposes the REPAIR framework to address noisy correspondence in cross-modal matching by utilizing rank correlation and memory banks.
