PEARLAlgorithm<T, TInput, TOutput>
Models & Types · AiDotNet.MetaLearning.Algorithms
Implementation of PEARL: Probabilistic Embeddings for Actor-critic RL (Rakelly et al., ICML 2019).
Inheritance: ModelBase<T, TInput, TOutput> → MetaLearnerBase<T, TInput, TOutput> → PEARLAlgorithm<T, TInput, TOutput>
Implements: IGeneratedParameterRegistrar<T>
How It Works
PEARL infers a latent task variable z from context (support data) using a probabilistic encoder q(z|c). The posterior is permutation invariant: a product of independent Gaussian factors, one per context example (paper Eq. 4), and z is sampled via reparameterization. The model is conditioned on z through parameter modulation (a generic model has no input slot for z). At test time, z is inferred from support data without gradient updates.
Algorithm:
Constructors
| Constructor | Summary |
|---|---|
PEARLAlgorithm(PEARLOptions<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) |