Fair Visual Recognition via Intervention with Proxy Features

Yi Zhang, Jitao Sang, Junyang Wang

Deep learning models often learn to make predictions that rely on sensitive social attributes like gender and race, which poses significant fairness risks, especially in societal a…