Class FederatedMetaLearningOptions

Namespace

AiDotNet.Models.Options

Assembly

AiDotNet.dll

Configuration options for federated meta-learning.

public sealed class FederatedMetaLearningOptions

Inheritance

object

FederatedMetaLearningOptions

Inherited Members

object.Equals(object)

object.Equals(object, object)

object.GetHashCode()

object.GetType()

object.ReferenceEquals(object, object)

object.ToString()

Remarks

For Beginners: Meta-learning in federated settings aims to learn a "good starting point" (initial model) that can adapt quickly to each client's local data with a small amount of fine-tuning.

In this library, federated meta-learning is implemented as an alternative server update rule that uses client adaptation results (post-local training) to update the global initialization.

Properties

Enabled

Gets or sets whether federated meta-learning is enabled.

public bool Enabled { get; set; }

Property Value

bool

InnerEpochs

Gets or sets the number of local adaptation epochs used for the inner loop.

public int InnerEpochs { get; set; }

Property Value

int

Remarks

If not set (or <= 0), the trainer falls back to the federated LocalEpochs value.

MetaLearningRate

Gets or sets the server meta learning rate applied to the average adaptation delta.

public double MetaLearningRate { get; set; }

Property Value

double

Remarks

For Beginners: This controls how strongly the server moves the global initialization toward the client-adapted models each round. A value of 1.0 means "move fully to the average adapted model" (similar to FedAvg when inner epochs match).

Strategy

Gets or sets the federated meta-learning strategy name.

public string Strategy { get; set; }

Property Value

string

Remarks

Supported built-ins:

• "None"
• "Reptile" (first-order meta-update based on post-adaptation parameters)
• "PerFedAvg" (treated as a Reptile-style first-order update in v1)
• "FedMAML" (first-order approximation in v1; full second-order requires explicit gradient support)