TimeVaryingTopology<T>
Models & Types · AiDotNet.FederatedLearning.Decentralized
Implements time-varying topology for decentralized federated learning.
Inheritance: FederatedLearningComponentBase<T> → TimeVaryingTopology<T>
Implements: IDecentralizedTopology
For Beginners
In decentralized FL, the communication graph determines which clients can talk to each other. A fixed graph can create “bottleneck” nodes and slow convergence. Time-varying topology changes the graph each round — this accelerates mixing (spreading information across all clients) and makes the system more robust to node failures.
How It Works
Strategies:
Reference: Time-Varying Communication Topologies for Decentralized FL (2024).
Constructors
| Constructor | Summary |
|---|---|
TimeVaryingTopology(TopologyStrategy<T> strategy, int seed) |
Creates a new time-varying topology. |
Properties
| Property | Summary |
|---|---|
TopologyStrategy<T> Strategy { get; } |
Gets the topology strategy. |
int RoundCounter { get; } |
Gets the current round. |
string TopologyName { get; } |
Methods
| Method | Summary |
|---|---|
Dictionary<int, List<int>> GenerateTopology(IReadOnlyList<int> clientIds) |
Generates the neighbor set for each client for the current round. |
double GetMixingWeight(int nodeId, int peerId, int totalNodes) |
|
int[] GetPeers(int nodeId, int totalNodes, int round) |