Loading AI Engineering Foundations...
A label is the ground truth target value or desired outcome associated with an example in supervised learning, against which model predictions are scored.
“The teacher’s answer key at the back of a math textbook.”
Low loss penalty! The model prediction aligns closely with the verified ground truth.
// Supervised dataset pair: [features, groundTruthLabel]
const sample = { features: [0.82, 0.45], label: 1 };Label noise: inconsistent or contradictory labels created by human raters degrade model ceiling accuracy.