Learn Glossary Blog Tools
AI Glossary

Label (Ground Truth)

The correct answer or target value associated with a training example in supervised learning, used to calculate loss and guide model training.

Types of Labels

Classification: Discrete categories (cat, dog, bird). Regression: Continuous values (price = $45.50). Sequence labeling: Per-token labels (NER tags). Bounding boxes: Object locations in images.

Annotation

Labels are created by human annotators, automated systems, or LLMs. Data annotation is often the most expensive and time-consuming part of ML projects. Platforms like Scale AI, Labelbox, and Amazon SageMaker Ground Truth help manage annotation.

Label Quality

Noisy or incorrect labels directly hurt model performance. Inter-annotator agreement metrics measure label quality. Techniques like label smoothing (softening hard labels) and confident learning help handle label noise.

← Back to AI Glossary

Related Articles

Infrastructure

Data Labeling Platforms: Building Training Datasets Efficiently

 Deep Learning

Image Segmentation: U-Net, Mask R-CNN, and SAM

 Deep Learning

Data Augmentation: Getting More from Less Data

 Computer Vision

Image Segmentation: Pixel-Level Understanding in AI

 Machine Learning

ML Model Evaluation: Accuracy, Precision, Recall, and F1

Related Concepts

Convolutional Neural Network (CNN) Data Annotation Data Augmentation Data Drift Data Leakage Data Pipeline Data Poisoning Image Segmentation
Last updated: March 5, 2026