DREAMAlgorithm<T, TInput, TOutput>
Models & Types · AiDotNet.MetaLearning.Algorithms
Implementation of DREAM: Directed REward Augmented Meta-learning.
Inheritance: ModelBase<T, TInput, TOutput> → MetaLearnerBase<T, TInput, TOutput> → DREAMAlgorithm<T, TInput, TOutput>
Implements: IGeneratedParameterRegistrar<T>
How It Works
DREAM meta-learns a reward shaping function that transforms the raw task loss into a more informative signal for the inner loop. The reward shaper is a small MLP that takes (loss, gradient_norm, step/K) as input and outputs a scalar multiplier for the gradient. This enables curriculum-like adaptation where early steps may be scaled differently than later steps.
Algorithm:
Constructors
| Constructor | Summary |
|---|---|
DREAMAlgorithm(DREAMOptions<T, TInput, TOutput> options) |
Properties
| Property | Summary |
|---|---|
MetaLearningAlgorithmType AlgorithmType { get; } |
Methods
| Method | Summary |
|---|---|
IModel<TInput, TOutput, ModelMetadata<T>> Adapt(IMetaLearningTask<T, TInput, TOutput> task) |
|
T MetaTrain(TaskBatch<T, TInput, TOutput> taskBatch) |