MetaBaselineAlgorithm<T, TInput, TOutput>

Models & Types · AiDotNet.MetaLearning.Algorithms

Implementation of Meta-Baseline (simple pre-train then meta-train with cosine classifier).

Inheritance: ModelBase<T, TInput, TOutput> → MetaLearnerBase<T, TInput, TOutput> → MetaBaselineAlgorithm<T, TInput, TOutput>

For Beginners

Meta-Baseline shows that the simplest approach can be the best:

  1. Train normally on many classes to get good features
  2. Switch to episodic training with cosine-distance centroids
  3. At test time, classify by nearest centroid (cosine distance)

How It Works

Meta-Baseline trains a feature extractor with standard classification, then fine-tunes with episodic training using cosine similarity nearest-centroid classification.

Reference: Chen, Y., Liu, Z., Xu, H., Darrell, T., & Wang, X. (2021). Meta-Baseline: Exploring Simple Meta-Learning for Few-Shot Learning. ICLR 2021.

Constructors

Constructor Summary
MetaBaselineAlgorithm(MetaBaselineOptions<T, TInput, TOutput> options) Initializes a new Meta-Baseline 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)