Model Cards

EE 641: A Computational Introduction to Deep Learning

A model card is a short document that travels with a trained model: what it is for, how well it works, and where it fails. The format comes from Mitchell et al., Model Cards for Model Reporting (2019). Your final project includes one as a graded deliverable.

Required Sections

Intended use — The task the model was trained for, and the settings it was and was not designed to handle.

Performance metrics — The metrics you report in the final report, restated with the evaluation set identified. Numbers, not adjectives.

Training data — Source, size, class balance, preprocessing. Anything about the data a user would need to judge whether the model transfers to their setting.

Limitations — Where performance degrades: input types, distribution shift, failure cases you observed. Be specific; “may not generalize” says nothing.

Fairness considerations — If the data involves people, whether performance differs across groups, and what you measured. If it does not, say so in one line.

Ethical implications — Consequences of plausible misuse or over-trust, at the scale of your actual model.

Skeleton

# Model Card: <model name>

## Intended Use


## Performance

| Metric | Evaluation set | Value |
|--------|----------------|-------|

## Training Data


## Limitations


## Fairness Considerations


## Ethical Implications