Class FedDynHeterogeneityCorrection<T>

Namespace

AiDotNet.FederatedLearning.Heterogeneity

Assembly

AiDotNet.dll

FedDyn-style dynamic regularization using a per-client drift accumulator.

public sealed class FedDynHeterogeneityCorrection<T> : FederatedHeterogeneityCorrectionBase<T>, IFederatedHeterogeneityCorrection<T>

Type Parameters

T

Numeric type.

Inheritance

object

FederatedLearningComponentBase<T>

FederatedHeterogeneityCorrectionBase<T>

FedDynHeterogeneityCorrection<T>

Implements

IFederatedHeterogeneityCorrection<T>

Inherited Members

object.Equals(object)

object.Equals(object, object)

object.GetHashCode()

object.GetType()

object.ReferenceEquals(object, object)

object.ToString()

Remarks

For Beginners: FedDyn reduces client drift by maintaining an extra per-client state that accumulates how the client tends to move away from the global model.

Constructors

FedDynHeterogeneityCorrection(double)

public FedDynHeterogeneityCorrection(double alpha = 0.01)

Parameters

alpha double

Methods

Correct(int, int, Vector<T>, Vector<T>, int)

Returns corrected client parameters to be used for aggregation.

public override Vector<T> Correct(int clientId, int roundNumber, Vector<T> globalParameters, Vector<T> localParameters, int localEpochs)

Parameters

clientId int

Client identifier.

roundNumber int

Round number (0-indexed).

globalParameters Vector<T>

Global parameter vector at the start of the round.

localParameters Vector<T>

Client-trained parameter vector.

localEpochs int

Local epochs used for training (proxy for local steps in simulation).

Returns

Vector<T>

Corrected parameters.

GetCorrectionName()

Gets the name of the correction method.

public override string GetCorrectionName()

Returns

string