Pre-Publication V3. A machine learning method where in which an AI model is trained on input examples paired with specific, known correct output answers. The model adjusts its parameters to minimize errors, allowing enabling it to predict or generate similar content when given outputs for previously unseen inputs.
Deliberation Summary:
(1) Clarify language to avoid reference to generation of similar content, on the view that a model may classify or score data without generating anything. (2) Consider whether training mechanism, example tasks, and the contrast with unsupervised and self-supervised methods can be addressed in the standards work.
Pre-Publication V2. A machine learning method where an AI is trained on "labeled data," input examples paired with specific, correct output answers. The model studies these examples, adjustsing its parameters to minimize errors, allowing it to accurately predict or generate similar content when given previously new, unseen inputs.
Deliberation Summary:
Wording clarification on unseen inputs.
Pre-Publication V1. A machine learning method where an AI is trained on "labeled data," input examples paired with specific, correct output answers. The model studies these examples, adjusting its parameters to minimize errors, allowing it to accurately predict or generate similar content when given new, unseen inputs.