CoreSetStrategy<T, TInput, TOutput>

Models & Types · AiDotNet.ActiveLearning.Strategies.Diversity

CoreSet strategy for active learning using geometric diversity.

Implements: IDiversityStrategy<T, TInput, TOutput>

For Beginners

CoreSet is a diversity-based strategy that aims to select samples that best represent the entire dataset. It’s like choosing a small set of “core” points that cover the data space well.

How It Works

How CoreSet Works:

Distance Metrics:

When to Use:

Reference: Sener and Savarese “Active Learning for Convolutional Neural Networks: A Core-Set Approach” (ICLR 2018)

Constructors

Constructor Summary
CoreSetStrategy() Initializes a new CoreSet strategy with Euclidean distance.
CoreSetStrategy(DistanceMetric distanceMetric, ActiveLearnerConfig<T> config) Initializes a new CoreSet strategy with specified distance metric.

Properties

Property Summary
string Description { get; }
string Name { get; }

Methods

Method Summary
Matrix<T> ComputeDistanceMatrix(IDataset<T, TInput, TOutput> dataset)
T ComputeDiversity(TInput sample1, TInput sample2)
Vector<T> ComputeDensityWeights(IDataset<T, TInput, TOutput> pool)
Vector<T> ComputeDiversityScores(IDataset<T, TInput, TOutput> unlabeledData, IDataset<T, TInput, TOutput> labeledData)
Vector<T> ComputeScores(IFullModel<T, TInput, TOutput> model, IDataset<T, TInput, TOutput> unlabeledPool)
Vector<T> GetFeatureRepresentation(TInput input)
int[] SelectDiverseSamples(IDataset<T, TInput, TOutput> unlabeledData, IDataset<T, TInput, TOutput> labeledData, int numSamples)
int[] SelectSamples(IFullModel<T, TInput, TOutput> model, IDataset<T, TInput, TOutput> unlabeledPool, int batchSize)
void InitializeLabeledFeatures(IDataset<T, TInput, TOutput> labeledData, IFullModel<T, TInput, TOutput> model) Sets the initial labeled features for distance computation.
void Reset()
void UpdateState(int[] newlyLabeledIndices, TOutput[] labels)