FederatedGraphOptions

Options & Configuration · AiDotNet.Models.Options

Configuration options for federated graph learning.

For Beginners

Graph FL extends standard federated learning to handle graph-structured data (social networks, molecular graphs, knowledge graphs). Each client holds a subgraph, and the server coordinates GNN training across all subgraphs. These options control the graph-specific aspects of training.

How It Works

Key decisions:

Constructors

Constructor Summary
FederatedGraphOptions()

Properties

Property Summary
CrossClientEdgeOptions CrossClientEdges { get; set; } Gets or sets cross-client edge handling options.
GraphFLMode Mode { get; set; } Gets or sets the graph FL mode (subgraph, node, link, or graph classification).
GraphPartitionStrategy PartitionStrategy { get; set; } Gets or sets the graph partition strategy.
PseudoNodeStrategy PseudoNodeStrategy { get; set; } Gets or sets the pseudo-node strategy for handling missing cross-client neighbors.
SubgraphSamplingOptions Sampling { get; set; } Gets or sets subgraph neighborhood sampling options.
bool UsePrototypeLearning { get; set; } Gets or sets whether to use prototype-based learning instead of full model sharing.
double LocalLearningRate { get; set; } Gets or sets the learning rate of each client’s local GNN gradient-descent steps.
double NeighborhoodPrivacyEpsilon { get; set; } Gets or sets the neighborhood privacy epsilon for LDP on topology queries.
int HiddenDimension { get; set; } Gets or sets the hidden layer dimension for GNN layers.
int LocalEpochs { get; set; } Gets or sets the number of full-batch local GNN training steps per client per round.
int NodeFeatureDimension { get; set; } Gets or sets the dimensionality of node feature vectors.
int NumGnnLayers { get; set; } Gets or sets the number of GNN message-passing layers.
int PrototypesPerClass { get; set; } Gets or sets the number of prototypes per class for prototype-based learning.
int? NumberOfPartitions { get; set; } Gets or sets the number of target partitions when using graph partitioning.