OptiGradTrustAggregationStrategy<T>
Models & Types · AiDotNet.FederatedLearning.Aggregators
Implements OptiGradTrust (Optimized Gradient Trust) aggregation strategy with historical reputation tracking.
Inheritance: FederatedLearningComponentBase<T> → AggregationStrategyBase<Dictionary<string, T[]>, T> → ParameterDictionaryAggregationStrategyBase<T> → OptiGradTrustAggregationStrategy<T>
For Beginners
OptiGradTrust builds on trust-based defenses like FLTrust by adding a historical reputation system. Each client maintains a trust score that is updated over multiple rounds. A client that consistently sends aligned, constructive updates builds a higher reputation, while one that repeatedly deviates gets downweighted. This makes the defense more resilient to adaptive attackers who behave honestly for a few rounds then suddenly attack.
How It Works
Trust update rule:
Reference: Optimized Gradient Trust Scoring for Federated Learning (2025).
Constructors
| Constructor | Summary |
|---|---|
OptiGradTrustAggregationStrategy(double momentum, double minReputation) |
Initializes a new instance of the OptiGradTrustAggregationStrategy class. |
Properties
| Property | Summary |
|---|---|
double MinReputation { get; } |
Gets the minimum reputation floor. |
double Momentum { get; } |
Gets the EMA momentum for reputation updates. |
Methods
| Method | Summary |
|---|---|
Dictionary<string, T[]> Aggregate(Dictionary<int, Dictionary<string, T[]>> clientModels, Dictionary<int, double> clientWeights) |
|
string GetStrategyName() |