IAdaptedMetaModel<T>

Interfaces · AiDotNet.Interfaces

Extended interface for meta-learning adapted models that carry task-specific adaptation state beyond backbone parameters. Enables downstream code to access algorithm-specific adapted representations for evaluation, analysis, or custom classification.

Known implementations: AdaptedMetaModel<T, TInput, TOutput>, NeuralProcessModel<T, TInput, TOutput>

For Beginners

When a meta-learner adapts to a new task, it often computes enriched feature representations beyond just the backbone model’s output. This interface lets you access those enriched features for custom use cases, like building a nearest-neighbor classifier on the adapted features instead of relying solely on the backbone’s predictions.

How It Works

Meta-learning algorithms like EPNet, MCL, SetFeat, and ConstellationNet compute adapted representations during task adaptation (e.g., propagated features, projected features, set-encoded class representations). This interface exposes those representations so that evaluation code, visualization tools, or custom classifiers can use them.

Properties

Property Summary
Vector<T> AdaptedSupportFeatures { get; } Gets the adapted support features computed during task adaptation.
Vector<T> ParameterModulationFactors { get; } Gets the parameter modulation factors computed during adaptation.