FairnessConstraintType
Enums · AiDotNet.Models.Options
Specifies the type of fairness constraint to enforce during federated learning.
For Beginners
Fairness constraints ensure the global model treats all client groups equitably. Without constraints, the model may perform well for majority clients but poorly for underrepresented groups (e.g., a rural hospital in a federation dominated by urban ones).
Fields
| Field | Summary |
|---|---|
FairnessConstraintType AgnosticFairness |
AFL (Agnostic FL): minimax optimization that is agnostic to test distribution. |
FairnessConstraintType DemographicParity |
Demographic parity: model predictions should be independent of group membership. |
FairnessConstraintType EqualOpportunity |
Equal opportunity: true positive rate should be equal across groups. |
FairnessConstraintType EqualizedOdds |
Equalized odds: true positive and false positive rates should be equal across groups. |
FairnessConstraintType FedFair |
FedFair: multi-objective optimization balancing accuracy, fairness, and efficiency via Pareto scalarization. |
FairnessConstraintType MinimaxFairness |
Minimax fairness: minimize the worst-case performance across all client groups. |
FairnessConstraintType None |
No fairness constraint applied. |
FairnessConstraintType QFairFederatedLearning |
q-FFL: parameterized fairness via power-mean. |
FairnessConstraintType TiltedERM |
TERM (Tilted ERM): smooth interpolation between average and worst-case optimization using a tilt parameter. |
int value__ |