Researchers reduce bias in AI models while preserving or improving accuracy

Machine-learning models can fail when they try to make predictions for individuals who were underrepresented in the datasets they were trained on.