BatchActiveLearner<T>
Models & Types · AiDotNet.ActiveLearning
Turns a per-sample informativeness (uncertainty) ranking into a diversity-aware batch selection, so the chosen batch covers the pool instead of collecting near-duplicate uncertain samples.
How It Works
Naive active learning takes the top-N most uncertain samples; those are frequently redundant (many
near-identical points from the same uncertain region), wasting labeling budget. This selects the
batch greedily by informativeness − diversityWeight · redundancy, where redundancy is a
sample’s maximum similarity to what is already chosen (BADGE / facility-location style). It layers on
top of any uncertainty strategy — the strategy supplies the scores, this makes the batch
diverse.
Constructors
| Constructor | Summary |
|---|---|
BatchActiveLearner(double diversityWeight) |
Creates a batch active learner. |
Methods
| Method | Summary |
|---|---|
ActiveLearningSelection<T> Select(Vector<T> informativeness, Matrix<T> representation, int batchSize, string strategyName, string representationSpace) |
Ranks the pool and selects a diversity-aware batch. |