FederatedContinualLearningOptions
Options & Configuration ยท AiDotNet.Models.Options
Configuration options for federated continual learning (preventing catastrophic forgetting in FL).
Inheritance: ModelOptions โ FederatedContinualLearningOptions
For Beginners
When federated models learn new tasks over time, they can forget what they learned before (catastrophic forgetting). These strategies identify which model parameters are important for previous tasks and protect them during future training rounds. Each client computes importance locally, and the server aggregates these importance estimates across all clients.
Constructors
| Constructor | Summary |
|---|---|
FederatedContinualLearningOptions() |
Properties
| Property | Summary |
|---|---|
FederatedContinualLearningStrategy Strategy { get; set; } |
Gets or sets the strategy. |
double DataFreeFCLDistillationTemperature { get; set; } |
Gets or sets the distillation temperature for data-free continual learning. |
double DataFreeFCLDistillationWeight { get; set; } |
Gets or sets the distillation loss weight for data-free FCL. |
double ExperienceReplayRatio { get; set; } |
Gets or sets the ratio of replay samples to new samples during training. |
double FedAGCCorrectionStrength { get; set; } |
Gets or sets the correction strength for adaptive gradient correction. |
double FedCILPrototypeDecay { get; set; } |
Gets or sets the EMA decay rate for prototype consolidation in FedCIL. |
double ProjectionThreshold { get; set; } |
Gets or sets the projection threshold for orthogonal projection. |
double RegularizationStrength { get; set; } |
Gets or sets the regularization strength for EWC penalty. |
int ExperienceReplayBufferCapacity { get; set; } |
Gets or sets the maximum buffer capacity for experience replay per client. |
int FisherSamples { get; set; } |
Gets or sets the number of data samples for Fisher information estimation. |