AdaptedMetaModel<T, TInput, TOutput>

Models & Types · AiDotNet.MetaLearning.Algorithms

Generic adapted model wrapper for meta-learning algorithms that use gradient-based inner-loop adaptation. After adaptation, the model uses the adapted parameters for prediction.

Inheritance: ModelWrapperBase<T, TInput, TOutput> → MetaLearningModelBase<T, TInput, TOutput> → AdaptedMetaModel<T, TInput, TOutput>

Implements: IAdaptedMetaModel<T>

For Beginners

An adapted meta-model is the result of a meta-learning algorithm (like MAML) adapting to a new task. After seeing a few examples of a new task, the meta-learner produces this adapted model with task-specific parameters. Think of it like a student who has learned general problem-solving skills and then quickly adapts to a specific exam topic after seeing just a few practice questions.

Example

var baseModel = new NeuralNetwork<float>(
    new NeuralNetworkArchitecture<float>(inputFeatures: 8, outputSize: 4));
            
// The parameters MAML's inner loop produced for one task.
var adaptedParams = new Vector<float>(baseModel.GetParameters().Length);
            
var trainX = Tensor<float>.CreateRandom(4, 8);
var trainY = Tensor<float>.CreateRandom(4, 4);
            
var result = new AiModelBuilder<float, Tensor<float>, Tensor<float>>()
    .ConfigureModel(new AdaptedMetaModel<float, Tensor<float>, Tensor<float>>(baseModel, adaptedParams))
    .Build(trainX, trainY);
            
var queryInput = Tensor<float>.CreateRandom(1, 8);
Tensor<float> prediction = result.Predict(queryInput);

Constructors

Constructor Summary
AdaptedMetaModel(IFullModel<T, TInput, TOutput> model, Vector<T> adaptedParams, Vector<T> supportFeatures, Vector<T> modulationFactors)

Properties

Property Summary
Vector<T> AdaptedSupportFeatures { get; }
Vector<T> ParameterModulationFactors { get; }

Methods

Method Summary
IFullModel<T, TInput, TOutput> WithParameters(Vector<T> parameters)
TOutput Predict(TInput input)