AI Dictionary
Ground Truth
Definition
Information provided by direct observation as opposed to information provided by inference.
Deep Dive
Ground Truth refers to the accurate, factual, and verified information or data used as a benchmark for evaluating and training machine learning models. It represents direct observations or measurements from the real world, as opposed to information derived from inference, prediction, or artificial intelligence systems. Essentially, it's the "correct answer" or the reliable standard against which an AI model's performance is measured.
Examples & Use Cases
- 1A dataset of images where each image is manually labeled by a human expert as containing a "cat," "dog," or "bird"
- 2In medical diagnosis, patient records where the presence or absence of a disease is confirmed by a panel of doctors, lab tests, or biopsies
- 3For autonomous driving systems, real-world sensor data (e.g., camera feeds, LiDAR) where objects like pedestrians, vehicles, and lane lines are meticulously annotated by human operators
Related Terms
Data LabelingSupervised LearningAnnotation