Model map

« Back to Glossary Index

Model map (Model Card)

Definition

Model cards are structured documents that provide key information about an AI model, similar to a nutrition label. They describe the purpose, area of application, performance values and limitations of a model in order to Transparency and create trust. In AI governance, model maps serve as binding evidence of a model's design, testing procedures and responsible use. They help developers, business managers and regulatory authorities to understand the strengths and weaknesses of a model and to assess risks. One Model map summarizes data provenance, performance measures and known risks to promote accountability in development and document compliance with the EU AI Regulation.

Why is a Model map important?

  • Promotes Transparency and trust through clear Documentation of purpose, data and performance.
  • Supports compliance with legal requirements (e.g. EU AI Regulation) and internal guidelines.
  • Helps developers and stakeholders to recognize and address risks and bias.

Components of a Model map

  • Name and version: Clearly identifies the model.
  • Purpose and use cases: Describes what the model is intended for and in which situations it may be used.
  • Training data and performance metrics: Explains what data was used and how well the model performs.
  • Restrictions and risks: Indicates known weaknesses, bias or situations in which the model is unreliable.

How model maps support AI governance

  • Ensure that each model is documented in a traceable manner.
  • Facilitate audits and compliance checks.
  • Create a basis for responsible decisions on the use or withdrawal of a model.
« Back to Glossary Index
administrator