FederatedPersonalizationOptions

Options & Configuration · AiDotNet.Models.Options

Configuration options for personalized federated learning (PFL).

How It Works

For Beginners: Personalization means each client can end up with a model that works better for its own data, while still learning shared knowledge from other clients.

This options class controls which personalization algorithm is used (FedPer, FedRep, Ditto, pFedMe, clustered, etc.) and the key hyperparameters for those algorithms.

Constructors

Constructor Summary
FederatedPersonalizationOptions()

Properties

Property Summary
FederatedPersonalizationStrategy Strategy { get; set; } Gets or sets the personalization strategy.
bool Enabled { get; set; } Gets or sets whether personalization is enabled.
double DittoLambda { get; set; } Gets or sets the Ditto regularization strength (lambda).
double FedBABUHeadFraction { get; set; } Gets or sets the fraction of parameters considered “head” in FedBABU.
double FedPACCalibrationWeight { get; set; } Gets or sets the calibration weight for FedPAC post-aggregation calibration.
double FedPACSimilarityThreshold { get; set; } Gets or sets the cosine similarity threshold for FedPAC prototype alignment.
double FedRoDMixingAlpha { get; set; } Gets or sets the mixing alpha between generic and personalized classifier in FedRoD.
double FedSelectMaskRegularization { get; set; } Gets or sets the mask regularization strength for FedSelect sparsity.
double FedSelectMaskThreshold { get; set; } Gets or sets the binary mask threshold for FedSelect parameter selection.
double KNNPerLambda { get; set; } Gets or sets the interpolation strength between model predictions and kNN predictions.
double PFedGateInitValue { get; set; } Gets or sets the initial gate value for pFedGate (0 = fully global, 1 = fully local).
double PFedGateLearningRate { get; set; } Gets or sets the gate learning rate for pFedGate.
double PFedMeBeta { get; set; } Gets or sets the server mixing weight (beta): w <- (1 - beta) w + beta * mean of the clients’ local models.
double PFedMeLearningRate { get; set; } Gets or sets the local learning rate (eta) of the update w <- w - eta * lambda * (w - theta).
double PFedMeMu { get; set; } Gets or sets the pFedMe regularization strength (lambda in Dinh et al.
double PFedMePersonalLearningRate { get; set; } Gets or sets the step size of the K inner steps on the personalized model.
double PersonalizedParameterFraction { get; set; } Gets or sets the fraction of parameters treated as “personalized” (not aggregated globally).
int ClusterCount { get; set; } Gets or sets the number of clusters used for clustered personalization.
int FedAGHNSharedDimension { get; set; } Gets or sets the shared dimension for adaptive gradient heterogeneous networks.
int FedBABULocalFineTuneEpochs { get; set; } Gets or sets the number of local fine-tune epochs after receiving the global model (FedBABU).
int FedCPNumExperts { get; set; } Gets or sets the number of conditional computation experts in FedCP.
int KNNPerK { get; set; } Gets or sets the number of nearest neighbors (k) for kNN-Per inference.
int LocalAdaptationEpochs { get; set; } Gets or sets the number of extra local adaptation epochs applied after receiving the aggregated global model.
int PFedMeInnerSteps { get; set; } Gets or sets the number of gradient steps (K) that approximately solve each client’s personalized problem.