FedGnnAggregationStrategy<T>
Models & Types · AiDotNet.FederatedLearning.Graph
GNN-aware federated aggregation that weights contributions by subgraph topology characteristics.
Inheritance: FederatedLearningComponentBase<T> → FedGnnAggregationStrategy<T>
Implements: IGraphAggregationStrategy<T>
For Beginners
Standard FedAvg weights clients by dataset size (number of samples). For graph FL, “dataset size” is ambiguous — is it nodes, edges, or labeled nodes? This strategy uses a composite weight based on graph topology:
How It Works
Formula: weight_i = alpha * E_i/E_total + beta * L_i/L_total + gamma * deg_i/deg_avg, where E = edges, L = labeled nodes, deg = average degree, and alpha+beta+gamma = 1.
Constructors
| Constructor | Summary |
|---|---|
FedGnnAggregationStrategy(double edgeWeight, double labelWeight, double degreeWeight) |
Initializes a new instance of FedGnnAggregationStrategy. |
Properties
| Property | Summary |
|---|---|
string StrategyName { get; } |
Methods
| Method | Summary |
|---|---|
Tensor<T> Aggregate(Dictionary<int, Tensor<T>> clientModels, Dictionary<int, ClientGraphStats> clientGraphStats) |