QueryByCommitteeStrategy<T, TInput, TOutput>

Models & Types · AiDotNet.ActiveLearning.Strategies.Committee

Query By Committee (QBC) strategy for active learning.

Implements: ICommitteeStrategy<T, TInput, TOutput>

For Beginners

QBC uses a “committee” of diverse models to identify samples where the models disagree most. High disagreement indicates uncertainty about the true label.

How It Works

How QBC Works:

Disagreement Measures:

When to Use:

Reference: Seung et al. “Query by Committee” (1992)

Constructors

Constructor Summary
QueryByCommitteeStrategy(IEnumerable<IFullModel<T, TInput, TOutput>> committee, CommitteeDisagreementMeasure disagreementMeasure, int numClasses, ActiveLearnerConfig<T> config) Initializes a new QBC strategy with a pre-built committee.
QueryByCommitteeStrategy(int numClasses) Initializes a new QBC strategy with an empty committee.

Properties

Property Summary
IReadOnlyList<IFullModel<T, TInput, TOutput>> Committee { get; }
string Description { get; }
string Name { get; }

Methods

Method Summary
T ComputeDisagreement(TInput input)
Vector<T> ComputeScores(IFullModel<T, TInput, TOutput> model, IDataset<T, TInput, TOutput> unlabeledPool)
int[] SelectSamples(IFullModel<T, TInput, TOutput> model, IDataset<T, TInput, TOutput> unlabeledPool, int batchSize)
void AddCommitteeMember(IFullModel<T, TInput, TOutput> model)
void RemoveCommitteeMember(IFullModel<T, TInput, TOutput> model)
void Reset()
void UpdateCommittee(IDataset<T, TInput, TOutput> trainingData)
void UpdateState(int[] newlyLabeledIndices, TOutput[] labels)