What are Model Cards?
Model cards are structured documentation frameworks for machine learning models, introduced by Google in 2019. They provide transparent information about a model's development, intended use, performance, and limitations - similar to nutrition labels for food products.
Standard Sections
Model Details
- Developer/organization
- Model date and version
- Model type and architecture
- Training procedures
- License and citation
Intended Use
- Primary intended uses
- Primary intended users
- Out-of-scope uses
Factors
- Relevant factors (demographics, domains)
- Evaluation factors
Metrics
- Model performance measures
- Decision thresholds
- Variation approaches
Evaluation Data
- Datasets used
- Motivation for selection
- Preprocessing steps
Training Data
- Dataset information
- If different from evaluation data
Quantitative Analyses
- Performance across groups
- Intersectional analysis
Ethical Considerations
- Potential risks
- Mitigations implemented
- Caveats and recommendations
Benefits
- Transparency and accountability
- Informed deployment decisions
- Bias and fairness awareness
- Regulatory compliance support
- Reproducibility
Related Concepts
- Data sheets for datasets
- System cards (for AI systems)
- AI factsheets