AsyncFederatedLearningOptions
Options & Configuration · AiDotNet.Models.Options
Configuration options for asynchronous federated learning (FedAsync / FedBuff).
Inheritance: ModelOptions → AsyncFederatedLearningOptions
How It Works
For Beginners: In synchronous federated learning, the server waits for a whole “round” of clients before updating the global model. In asynchronous federated learning, the server can update as client updates arrive (or in small buffers), which can reduce waiting on slow clients.
Constructors
| Constructor | Summary |
|---|---|
AsyncFederatedLearningOptions() |
Properties
| Property | Summary |
|---|---|
FederatedAsyncMode Mode { get; set; } |
Gets or sets the async mode. |
FederatedStalenessWeighting StalenessWeighting { get; set; } |
Gets or sets the staleness weighting mode for asynchronous updates. |
double AsyncFedEDExplorationBonus { get; set; } |
Gets or sets the exploration bonus for unvisited clients in AsyncFedED scheduling. |
double FedAsyncMixingRate { get; set; } |
Gets or sets the base mixing rate used by FedAsync. |
double StalenessDecayRate { get; set; } |
Gets or sets the staleness decay rate used by “Exponential” and “Polynomial” weighting. |
int AsyncFedEDSelectionBudget { get; set; } |
Gets or sets the maximum clients to select per round in AsyncFedED. |
int FedBuffBufferSize { get; set; } |
Gets or sets the buffer size for FedBuff (number of updates to accumulate before applying a server update). |
int RejectUpdatesWithStalenessGreaterThan { get; set; } |
Gets or sets the maximum staleness allowed before rejecting an update. |
int SemiAsyncRoundsPerBarrier { get; set; } |
Gets or sets the number of async rounds between synchronization barriers (Semi-Async mode). |
int SimulatedMaxClientDelaySteps { get; set; } |
Gets or sets the maximum simulated client delay (in server steps) for in-memory async training. |