active learning for inspection

**Active Learning for Inspection** is a **strategy where the ML model selectively requests labels for the most informative samples** — minimizing the total labeling effort by intelligently choosing which defect images to send to human experts for annotation. **How Active Learning Works** - **Initial Model**: Train a model on a small initial labeled set. - **Query Strategy**: Select the most uncertain or informative unlabeled samples for labeling. - **Human Label**: Expert annotates only the selected samples. - **Retrain**: Update the model with newly labeled data, repeat. - **Strategies**: Uncertainty sampling, query-by-committee, diversity sampling. **Why It Matters** - **Label Efficiency**: Achieves target accuracy with 50-80% fewer labeled samples compared to random labeling. - **Expert Time**: Fab defect labeling requires expensive domain experts — active learning minimizes their workload. - **Evolving Distribution**: Continuously adapts to new defect types by requesting labels for unknown patterns. **Active Learning** is **smart labeling for defect inspection** — letting the AI ask the expert about the most confusing samples to learn faster with less labeling.