IFairnessConstraint<T>
Interfaces · AiDotNet.FederatedLearning.Fairness
Defines and enforces fairness constraints during federated learning aggregation.
Known implementations: GroupFairnessConstraint<T>
For Beginners
A fairness constraint ensures the global model works well for ALL client groups, not just the majority. Without it, a model trained across 10 urban hospitals and 2 rural hospitals might be great for urban patients but poor for rural ones. Fairness constraints rebalance the aggregation to protect underrepresented groups.
How It Works
Common fairness measures:
Properties
| Property | Summary |
|---|---|
string ConstraintName { get; } |
Gets the name of this fairness constraint. |
Methods
| Method | Summary |
|---|---|
Dictionary<int, double> AdjustWeights(Dictionary<int, double> originalWeights, Dictionary<int, Tensor<T>> clientModels, Tensor<T> globalModel, Dictionary<int, int> clientGroups) |
Adjusts aggregation weights to enforce fairness constraints. |
bool IsSatisfied(double fairnessViolation) |
Gets whether the current model satisfies the fairness constraint. |
double EvaluateFairness(Dictionary<int, Tensor<T>> clientModels, Tensor<T> globalModel, Dictionary<int, int> clientGroups) |
Evaluates the current fairness metric across client groups. |