NPBMLAlgorithm<T, TInput, TOutput>
Models & Types · AiDotNet.MetaLearning.Algorithms
Implementation of NPBML (Neural Process-Based Meta-Learning).
Inheritance: ModelBase<T, TInput, TOutput> → MetaLearnerBase<T, TInput, TOutput> → NPBMLAlgorithm<T, TInput, TOutput>
Implements: IGeneratedParameterRegistrar<T>
For Beginners
NPBML is a probabilistic meta-learner that knows when it’s uncertain:
Standard meta-learners: Given a support set, produce ONE adapted model that gives ONE prediction per query. You don’t know how confident the model is.
NPBML’s probabilistic approach:
- Encode support examples into a DISTRIBUTION (mean + variance), not a single point
- Sample from this distribution multiple times
- Each sample gives a slightly different prediction
- If samples agree: model is confident
- If samples disagree: model is uncertain about this task
Why this matters:
- Few support examples = high uncertainty (wide distribution)
- Many similar support examples = low uncertainty (narrow distribution)
- Ambiguous task = high uncertainty (can’t determine the right adaptation)
Analogy: It’s like asking 5 experts (sampled from the distribution). If all 5 agree, you’re confident. If they give different answers, you know the task is ambiguous.
How It Works
NPBML combines neural processes with meta-learning for probabilistic few-shot prediction. It encodes the support set into a latent distribution and samples from it to capture task-level uncertainty.
Algorithm - NPBML:
Constructors
| Constructor | Summary |
|---|---|
NPBMLAlgorithm(NPBMLOptions<T, TInput, TOutput> options) |
Initializes a new NPBML meta-learner. |
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) |