#data distribution shifts

Google researchers developed a causal framework that improves the reliability of subgroup fairness evaluations in machine learning by accounting for data distribution and structural biases.