FederatedLearningOptions

Options & Configuration · AiDotNet.Models.Options

Configuration options for federated learning training.

Inheritance: ModelOptions → FederatedLearningOptions

How It Works

This class contains all the configurable parameters needed to set up and run a federated learning system.

For Beginners: Options are like the settings panel for federated learning. Just as you configure settings for a video game (difficulty, graphics quality, etc.), these options let you configure how federated learning should work.

Key configuration areas:

  • Client Management: How many clients, how to select them
  • Training: Learning rates, epochs, batch sizes
  • Privacy: Differential privacy parameters
  • Communication: How often to aggregate, compression settings
  • Convergence: When to stop training

For example, a typical configuration might be:

  • 100 total clients (e.g., hospitals)
  • Select 10 clients per round (10% participation)
  • Each client trains for 5 local epochs
  • Use privacy budget ε=1.0, δ=1e-5
  • Run for maximum 100 rounds or until convergence

Constructors

Constructor Summary
FederatedLearningOptions()

Properties

Property Summary
AdvancedCompressionOptions AdvancedCompression { get; set; } Gets or sets advanced communication compression options (PowerSGD, sketching, error feedback).
AsyncFederatedLearningOptions AsyncFederatedLearning { get; set; } Gets or sets asynchronous federated learning options (FedAsync / FedBuff).
BackdoorDefenseOptions BackdoorDefense { get; set; } Gets or sets backdoor attack detection and defense options.
ClientSelectionOptions ClientSelection { get; set; } Gets or sets client selection options (strategy and related parameters).
ContributionEvaluationOptions ContributionEvaluation { get; set; } Gets or sets client contribution evaluation options for measuring each client’s value.
DecentralizedFederatedOptions Decentralized { get; set; } Gets or sets decentralized (serverless) federated learning options.
DifferentialPrivacyMode DifferentialPrivacyMode { get; set; } Gets or sets where differential privacy is applied (local, central, or both).
FederatedAdapterOptions Adapters { get; set; } Gets or sets federated adapter options for parameter-efficient fine-tuning (LoRA, prompt tuning).
FederatedAggregationStrategy AggregationStrategy { get; set; } Gets or sets the aggregation strategy to use.
FederatedCompressionOptions Compression { get; set; } Gets or sets federated compression options.
FederatedContinualLearningOptions ContinualLearning { get; set; } Gets or sets federated continual learning options for preventing catastrophic forgetting.
FederatedDistillationOptions Distillation { get; set; } Gets or sets federated knowledge distillation options (FedMD, FedDF, FedGEN).
FederatedDriftOptions DriftDetection { get; set; } Gets or sets federated concept drift detection and adaptation options.
FederatedFairnessOptions Fairness { get; set; } Gets or sets fairness constraint options for equitable model performance across client groups.
FederatedGraphOptions GraphLearning { get; set; } Gets or sets federated graph learning options.
FederatedHeterogeneityCorrectionOptions HeterogeneityCorrection { get; set; } Gets or sets federated heterogeneity correction options (SCAFFOLD / FedNova / FedDyn).
FederatedLearningMode Mode { get; set; } Gets or sets the federated learning mode (horizontal or vertical).
FederatedMetaLearningOptions MetaLearning { get; set; } Gets or sets federated meta-learning options (Per-FedAvg / FedMAML / Reptile-style).
FederatedPersonalizationOptions Personalization { get; set; } Gets or sets personalization options (preferred configuration for FedPer/FedRep/Ditto/pFedMe/clustered).
FederatedPrivacyAccountant PrivacyAccountant { get; set; } Gets or sets which privacy accountant to use for reporting privacy spend.
FederatedServerOptimizerOptions ServerOptimizer { get; set; } Gets or sets server-side federated optimization options (FedOpt family).
FederatedUnlearningOptions Unlearning { get; set; } Gets or sets federated unlearning options (right to be forgotten).
HomomorphicEncryptionOptions HomomorphicEncryption { get; set; } Gets or sets homomorphic encryption options for federated aggregation (CKKS/BFV).
MpcOptions MultiPartyComputation { get; set; } Gets or sets Multi-Party Computation options for secure operations beyond summation.
PsiOptions PrivateSetIntersection { get; set; } Gets or sets Private Set Intersection options for entity alignment.
RobustAggregationOptions RobustAggregation { get; set; } Gets or sets options for robust aggregation strategies (Median, TrimmedMean, Krum, MultiKrum, Bulyan).
SecureAggregationOptions SecureAggregation { get; set; } Gets or sets secure aggregation configuration options.
TeeOptions TrustedExecutionEnvironment { get; set; } Gets or sets Trusted Execution Environment options for hardware-backed secure aggregation.
VerificationOptions Verification { get; set; } Gets or sets verification options for zero-knowledge proof-based update validation.
VerticalFederatedLearningOptions VerticalLearning { get; set; } Gets or sets vertical federated learning options (split learning, entity alignment, etc.).
bool EnablePersonalization { get; set; } Gets or sets whether to enable personalization.
bool UseCompression { get; set; } Gets or sets whether to use gradient compression to reduce communication costs.
bool UseDifferentialPrivacy { get; set; } Gets or sets whether to use differential privacy.
bool UseSecureAggregation { get; set; } Gets or sets whether to use secure aggregation.
double ClientSelectionFraction { get; set; } Gets or sets the fraction of clients to select for each training round (0.0 to 1.0).
double CompressionRatio { get; set; } Gets or sets the compression ratio (0.0 to 1.0) if compression is enabled.
double ConvergenceThreshold { get; set; } Gets or sets the convergence threshold for early stopping.
double DifferentialPrivacyClipNorm { get; set; } Gets or sets the clipping norm used by differential privacy mechanisms.
double LearningRate { get; set; } Gets or sets the learning rate for local client training.
double PersonalizationLayerFraction { get; set; } Gets or sets the fraction of model layers to keep personalized (not aggregated).
double PrivacyDelta { get; set; } Gets or sets the delta (δ) parameter for differential privacy (failure probability).
double PrivacyEpsilon { get; set; } Gets or sets the epsilon (ε) parameter for differential privacy (privacy budget).
double ProximalMu { get; set; } Gets or sets the proximal term coefficient for FedProx algorithm.
int BatchSize { get; set; } Gets or sets the batch size for local training.
int LocalEpochs { get; set; } Gets or sets the number of local training epochs each client performs per round.
int MaxRounds { get; set; } Gets or sets the maximum number of federated learning rounds to execute.
int MinRoundsBeforeConvergence { get; set; } Gets or sets the minimum number of rounds to train before checking convergence.
int NumberOfClients { get; set; } Gets or sets the total number of clients participating in federated learning.