Kartenstapel

User-to-Data Bias

Data-to-Algorithm Bias

Algorithm-to-User Bias

Human Cognitive Bias

Data Bias

Engineering Bias

Individual Fairness: Similar predictions for similar individuals -> Ensure that model treates all individuals equaly regardless of characteristics like race or gender

Group Fairness: Equal treatment of different groups when group membership is not causal to treatment -> Ensure that models predictions does not disproportionately affect certain groups that are defined on protected attributes like race or gender

Subgroup Fairness: Checks whether fairness criteria hold over several subgroups -> Ensure that models prediciton is are fair for all subgroups of different groups in the population, such as specific demographic groups