In order to perform supervised learning, we need some form of ground truth, that is for at least some training data we need to know the correct classification or output value. This may be obtained through human labelling of data, from sensor data or in some cases simulation.
Used in glossary entries: human labelling, supervised learning, training data