Training data

Training data shapes what the model learns, including its biases. Keeping it separate from tuning and test data, and documenting where it came from, is central to an AI device submission.

Article 3 (29) AIA defines ‘training data’ as data used for training an AI system through fitting its learnable parameters;

MDCG 2025-6 / AIB 2025-1, Section II, Question 10 (citing Article 3(29) AIA)

These data are used by the manufacturer of an AI-DSF in procedures and training algorithms to build an AI model, including to define model weights, connections, and components. These data typically should be representative of the proposed intended use populations (e.g., with respect to race, ethnicity, disease severity, sex, age, or others, as appropriate) and intended environments.

FDA guidance: Marketing Submission Recommendations for a Predetermined Change Control Plan for Artificial Intelligence-Enabled Device Software Functions (August 2025), Section IV, Definitions

The AI Act definition quoted in MDCG 2025-6 covers data used to fit an AI system's learnable parameters. The FDA guidance describes data the manufacturer uses to build the model, including its weights, connections and components, and adds that it should typically be representative of the intended use population.