PrototypicalContributionEvaluator<T>
Models & Types · AiDotNet.FederatedLearning.Fairness
Prototypical contribution evaluator: measures client value using prototype representations.
Inheritance: FederatedLearningComponentBase<T> → PrototypicalContributionEvaluator<T>
Implements: IClientContributionEvaluator<T>
For Beginners
Instead of expensive Shapley value computation (which requires testing many model combinations), this method uses a shortcut: compare what each client “knows” (represented as prototype vectors — like a summary of their data) against the global model. Clients whose prototypes align well with the global direction but also bring unique information score highest.
How It Works
How it works:
Advantage: Constant cost per round O(N * D), where N = clients, D = model size. Much cheaper than Shapley for large federations.
Constructors
| Constructor | Summary |
|---|---|
PrototypicalContributionEvaluator(ContributionEvaluationOptions options) |
Initializes a new instance of PrototypicalContributionEvaluator. |
Properties
| Property | Summary |
|---|---|
string MethodName { get; } |
Methods
| Method | Summary |
|---|---|
Dictionary<int, double> EvaluateContributions(Dictionary<int, Tensor<T>> clientModels, Tensor<T> globalModel, Dictionary<int, List<Tensor<T>>> clientHistories) |
|
HashSet<int> IdentifyFreeRiders(Dictionary<int, double> contributionScores) |