Overview
Direct Answer
Active learning is a machine learning paradigm in which an algorithm selectively queries an oracle (typically a human annotator) to label the most informative unlabelled data points, rather than passively consuming a pre-labelled dataset. This approach reduces annotation effort whilst maintaining or improving model performance.
How It Works
The algorithm trains on an initial small labelled set, then iteratively identifies which unlabelled samples would provide the greatest reduction in model uncertainty or error if annotated. Selection strategies include uncertainty sampling (highest entropy predictions), query-by-committee (disagreement among ensemble members), and expected model change. The newly labelled samples are incorporated into the training set, and the process repeats until a stopping criterion is met.
Why It Matters
Organisations face significant costs when acquiring expert labels, particularly in domains requiring domain-specific knowledge such as medical imaging, compliance review, or scientific research. Active learning can reduce labelling costs by 50–80 per cent relative to random sampling whilst achieving equivalent model accuracy, accelerating deployment timelines and reducing expenses for resource-constrained teams.
Common Applications
Applications include medical diagnosis systems where radiologist annotations are expensive, sentiment analysis in low-resource languages, anomaly detection in security systems, and biological sequence classification. Legal technology firms employ active learning to optimise document review workflows by prioritising uncertain cases for human review.
Key Considerations
The effectiveness of active learning depends heavily on the quality of the selection strategy and the availability of reliable oracles; poor query design can waste annotations. Additionally, active learning introduces complexity in model validation and may exhibit suboptimal performance in highly imbalanced datasets or when the initial sample is unrepresentative.
Cross-References(2)
More in Machine Learning
Transfer Learning
Advanced MethodsA technique where knowledge gained from training on one task is applied to a different but related task.
Gradient Boosting
Supervised LearningAn ensemble technique that builds models sequentially, with each new model correcting residual errors of the combined ensemble.
Polynomial Regression
Supervised LearningA form of regression analysis where the relationship between variables is modelled as an nth degree polynomial.
Catastrophic Forgetting
Anomaly & Pattern DetectionThe tendency of neural networks to completely lose previously learned knowledge when trained on new tasks, a fundamental challenge in continual and multi-task learning.
Cross-Validation
Training TechniquesA resampling technique that partitions data into subsets, training on some and validating on others to assess model generalisation.
Ensemble Methods
MLOps & ProductionMachine learning techniques that combine multiple models to produce better predictive performance than any single model, including bagging, boosting, and stacking approaches.
Bandit Algorithm
Advanced MethodsAn online learning algorithm that balances exploration of new options with exploitation of known good options to maximise reward.
Ridge Regression
Training TechniquesA regularised regression technique that adds an L2 penalty term to prevent overfitting by constraining coefficient magnitudes.