FedAaAggregationStrategy<T>

Models & Types · AiDotNet.FederatedLearning.Aggregators

Implements FedAA (Federated Adaptive Aggregation) strategy.

Inheritance: FederatedLearningComponentBase<T> → AggregationStrategyBase<Dictionary<string, T[]>, T> → ParameterDictionaryAggregationStrategyBase<T> → FedAaAggregationStrategy<T>

For Beginners

In standard FedAvg, each client’s contribution is weighted only by sample count. FedAA learns better aggregation weights by measuring how similar each client’s update direction is to the overall update direction. Clients whose updates are more “aligned” with the consensus get higher weight, while outlier updates get lower weight.

How It Works

Weight computation:

Reference: Adaptive Aggregation for Federated Learning (2024).

Constructors

Constructor Summary
FedAaAggregationStrategy(double temperature) Initializes a new instance of the FedAaAggregationStrategy class.

Properties

Property Summary
double Temperature { get; } Gets the softmax temperature for attention weighting.

Methods

Method Summary
Dictionary<string, T[]> Aggregate(Dictionary<int, Dictionary<string, T[]>> clientModels, Dictionary<int, double> clientWeights)
string GetStrategyName()