Table of Contents

Class MAMLAlgorithm<T, TInput, TOutput>

Namespace
AiDotNet.MetaLearning.Algorithms
Assembly
AiDotNet.dll

Implementation of the MAML (Model-Agnostic Meta-Learning) algorithm.

public class MAMLAlgorithm<T, TInput, TOutput> : MetaLearnerBase<T, TInput, TOutput>, IMetaLearner<T, TInput, TOutput>

Type Parameters

T

The numeric type used for calculations (e.g., double, float).

TInput

The input data type (e.g., Matrix<T>, Tensor<T>).

TOutput

The output data type (e.g., Vector<T>, Tensor<T>).

Inheritance
MetaLearnerBase<T, TInput, TOutput>
MAMLAlgorithm<T, TInput, TOutput>
Implements
IMetaLearner<T, TInput, TOutput>
Inherited Members

Remarks

MAML (Model-Agnostic Meta-Learning) is a meta-learning algorithm that trains models to be easily fine-tunable. It learns initial parameters such that a small number of gradient steps on a new task will lead to good performance.

Key features: - Model-agnostic: works with any model trainable with gradient descent - Learns good initialization rather than learning a fixed feature extractor - Enables few-shot learning with just 1-5 examples per class

For Beginners: MAML is like teaching someone how to learn quickly.

Normal machine learning: Train a model for one specific task MAML: Train a model to be easily trainable for many different tasks

It's like learning how to learn - by practicing on many tasks, the model learns what kind of parameters make it easy to adapt to new tasks quickly.

Reference: Finn, C., Abbeel, P., & Levine, S. (2017). Model-agnostic meta-learning for fast adaptation of deep networks.

Constructors

MAMLAlgorithm(MAMLOptions<T, TInput, TOutput>)

Initializes a new instance of the MAMLAlgorithm class.

public MAMLAlgorithm(MAMLOptions<T, TInput, TOutput> options)

Parameters

options MAMLOptions<T, TInput, TOutput>

MAML configuration options containing the model and all hyperparameters.

Examples

// Create MAML with minimal configuration (uses all defaults)
var options = new MAMLOptions<double, Tensor, Tensor>(myNeuralNetwork);
var maml = new MAMLAlgorithm<double, Tensor, Tensor>(options);

// Create MAML with custom configuration
var options = new MAMLOptions<double, Tensor, Tensor>(myNeuralNetwork)
{
    LossFunction = new CrossEntropyLoss<double>(),
    InnerLearningRate = 0.01,
    OuterLearningRate = 0.001,
    AdaptationSteps = 5,
    UseFirstOrderApproximation = true
};
var maml = new MAMLAlgorithm<double, Tensor, Tensor>(options);

Exceptions

ArgumentNullException

Thrown when options is null.

InvalidOperationException

Thrown when required components are not set in options.

Properties

AlgorithmType

Gets the algorithm type identifier for this meta-learner.

public override MetaLearningAlgorithmType AlgorithmType { get; }

Property Value

MetaLearningAlgorithmType

Returns MAML.

Remarks

This property identifies the algorithm as MAML (Model-Agnostic Meta-Learning), which is useful for serialization, logging, and algorithm-specific handling.

Methods

Adapt(IMetaLearningTask<T, TInput, TOutput>)

Adapts the meta-learned model to a new task using MAML's inner loop optimization.

public override IModel<TInput, TOutput, ModelMetadata<T>> Adapt(IMetaLearningTask<T, TInput, TOutput> task)

Parameters

task IMetaLearningTask<T, TInput, TOutput>

The new task containing support set examples for adaptation.

Returns

IModel<TInput, TOutput, ModelMetadata<T>>

A new model instance that has been fine-tuned to the given task.

Remarks

This is where MAML's "learning to learn" capability shines. The meta-learned initialization allows the model to quickly adapt to new tasks with just a few gradient steps on a small support set.

Adaptation Process: 1. Clone the meta-model to preserve the learned initialization 2. Perform K gradient descent steps on the task's support set 3. Return the adapted model, ready for inference on new examples

For Beginners: After meta-training, call this method when you have a new task with a few labeled examples. The returned model will be specialized for that task and ready to make predictions.

The number of gradient steps is controlled by AdaptationSteps. Typically 1-10 steps are sufficient thanks to the good initialization learned during meta-training.

Exceptions

ArgumentNullException

Thrown when task is null.

MetaTrain(TaskBatch<T, TInput, TOutput>)

Performs one meta-training step using MAML's bi-level optimization.

public override T MetaTrain(TaskBatch<T, TInput, TOutput> taskBatch)

Parameters

taskBatch TaskBatch<T, TInput, TOutput>

A batch of tasks to meta-train on, each containing support and query sets.

Returns

T

The average meta-loss across all tasks in the batch.

Remarks

MAML meta-training consists of two nested optimization loops:

Inner Loop (Task Adaptation): For each task in the batch: 1. Clone the meta-model with current meta-parameters 2. Perform K gradient descent steps on the task's support set 3. Evaluate the adapted model on the task's query set

Outer Loop (Meta-Update): 1. Compute meta-gradients based on query set performance 2. Average meta-gradients across all tasks in the batch 3. Apply gradient clipping if configured 4. Update meta-parameters using the averaged meta-gradients

For Beginners: Each call to this method makes the model slightly better at learning new tasks quickly. The returned loss value should decrease over time.

Exceptions

ArgumentException

Thrown when the task batch is null or empty.

InvalidOperationException

Thrown when meta-gradient computation fails.

Edit this page