Machine learningOther

Arbitrariness and Social Prediction: The Confounding Role of Variance in Fair Classification

Variance in predictions across different models is a major source of error in fair binary classification, and a new metric, self-consistency, is proposed to measure and reduce randomness, challenging the effectiveness of common algorithmic fairness methods.