VERSAOptions<T, TInput, TOutput>

Options & Configuration · AiDotNet.MetaLearning.Options

Configuration options for VERSA (Versatile and Efficient Few-shot Learning) algorithm.

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

Implements: IMetaLearnerOptions<T>

For Beginners

VERSA takes a completely different approach to few-shot learning:

Traditional meta-learning (MAML, etc.): “Here are 5 examples. Let me run gradient descent to learn a classifier…” (slow)

VERSA: “Here are 5 examples. single forward pass Here’s your classifier.” (instant)

How? VERSA trains a separate neural network (the “amortization network”) whose job is to look at a set of examples and immediately output the weights for a classifier. It’s like having a factory that produces customized classifiers on demand.

Key advantages:

  • No inner-loop optimization at all (fastest possible adaptation)
  • Naturally handles variable numbers of support examples
  • The amortization network learns to extract task-relevant statistics

How It Works

VERSA uses an amortization network to produce task-specific parameters in a single forward pass, eliminating the need for inner-loop gradient descent entirely. Given support examples, the amortization network directly outputs classifier parameters.

Reference: Gordon, J., Bronskill, J., Bauer, M., Nowozin, S., & Turner, R. E. (2019). Meta-Learning Probabilistic Inference for Prediction. ICLR 2019.

Constructors

Constructor Summary
VERSAOptions(IFullModel<T, TInput, TOutput> metaModel) Initializes a new instance of VERSAOptions.

Properties

Property Summary
IEpisodicDataLoader<T, TInput, TOutput> DataLoader { get; set; } Gets or sets the episodic data loader.
IFullModel<T, TInput, TOutput> MetaModel { get; set; } Gets or sets the feature extractor model.
IGradientBasedOptimizer<T, TInput, TOutput> InnerOptimizer { get; set; } Must stay null: this algorithm takes no plain gradient inner step (it adapts without one, or by its own update rule), so a supplied optimizer is rejected when the learner is built.
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; }
bool UseProbabilistic { get; set; } Gets or sets whether to use a probabilistic (Bayesian) formulation.
double AmortizationDropout { get; set; } Gets or sets the dropout rate for the amortization network.
double InnerLearningRate { get; set; }
double OuterLearningRate { get; set; }
double? GradientClipThreshold { get; set; }
int AdaptationSteps { get; set; }
int AmortizationHiddenDim { get; set; } Gets or sets the hidden dimension of the amortization network.
int AmortizationNumLayers { get; set; } Gets or sets the number of hidden layers in the amortization network.
int CheckpointFrequency { get; set; }
int EvaluationFrequency { get; set; }
int EvaluationTasks { get; set; }
int MetaBatchSize { get; set; }
int NumMetaIterations { get; set; }
int? RandomSeed { get; set; } Gets or sets the random seed.
string AggregationMethod { get; set; } Gets or sets the aggregation method for support set features.

Methods

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