PACOHOptions<T, TInput, TOutput>

Options & Configuration · AiDotNet.MetaLearning.Options

Configuration options for PACOH: PAC-Bayesian Meta-Learning with Optimal Hyperparameters (Rothfuss et al., ICLR 2021).

Inheritance: ModelOptions → PACOHOptions<T, TInput, TOutput>

Implements: IMetaLearnerOptions<T>

How It Works

PACOH meta-learns a Gaussian prior N(μ, σ²I) over neural network parameters that provides PAC-Bayesian generalization guarantees. The outer loop optimizes the prior to minimize a PAC-Bayesian bound; the inner loop performs MAP estimation with the learned prior as regularizer.

Constructors

Constructor Summary
PACOHOptions(IFullModel<T, TInput, TOutput> metaModel) Creates options for meta-learning metaModel, with the default settings.

Properties

Property Summary
IEpisodicDataLoader<T, TInput, TOutput> DataLoader { get; set; } Episodic data loader that supplies the meta-training tasks; null when tasks are passed to training directly.
IFullModel<T, TInput, TOutput> MetaModel { get; set; } The model whose parameters the meta-learner trains to adapt quickly to new tasks.
IGradientBasedOptimizer<T, TInput, TOutput> InnerOptimizer { get; set; } Optimizer for task adaptation.
IGradientBasedOptimizer<T, TInput, TOutput> MetaOptimizer { get; set; } Optimizer for the meta-update.
ILossFunction<T> LossFunction { get; set; } Loss function for meta-training; null uses the default multi-class classification loss.
bool EnableCheckpointing { get; set; }
bool UseFirstOrder { get; set; } Whether to use the first-order approximation of the meta-gradient.
double Delta { get; set; } Confidence parameter δ for the PAC-Bayesian bound (higher = tighter but more conservative).
double InitialLogVariance { get; set; } Initial log-variance of the prior distribution.
double InnerLearningRate { get; set; }
double KLCoefficient { get; set; } KL divergence coefficient in the PAC-Bayesian bound.
double OuterLearningRate { get; set; }
double? GradientClipThreshold { get; set; }
int AdaptationSteps { get; set; }
int CheckpointFrequency { get; set; }
int EvaluationFrequency { get; set; }
int EvaluationTasks { get; set; }
int MetaBatchSize { get; set; }
int NumMetaIterations { get; set; }
int NumPosteriorSamples { get; set; } Number of posterior samples for Monte Carlo estimation of expected loss.
int? RandomSeed { get; set; }

Methods

Method Summary
IMetaLearnerOptions<T> Clone()
bool IsValid()