The use and adoption of A.I.-integrated solutions and Machine Learning models has not always come hand-in-hand with efforts to increase their transparency. Model Cards aim to encapsulate all the relevant information needed to complete an audit in a succinct, easy-to-understand format.
In addition to greatly facilitating the model governance/risk and compliance approval process, Model Cards can encourage model democratization and reproducibility across the organization by detailing “the recipe” within the model itself rather than containing it with an individual or group of individuals.
We recommend providing the following type of information for a good Model Card:
- Model ownership. Which groups within the organization will leverage the model?
- Intended use/scope. Loan risk assessment, forecasting, anomaly detection, trade break classification)
- Ethical considerations. Could the model’s predictions seriously impact people’s lives?
- Evaluation metrics. Accuracy, f-score, recall, precision
- Required libraries and their corresponding versions. Has compliance checked the licensing of these?
- Training/creation date
- Hyper-parameters used in training: Can we recreate the results?
- Overview of training data. Should include: target labels, number of records, features used, etc.
EZOPS’ Model Cards Expedite the Approval of Models
We embedded Model Cards into the EZOPS A.I. Control Hub. Each model trained in our application can generate the requisite Model Card with a simple click of a button. Anyone with permissions to the model can quickly export the report on demand as a pdf and access the information contained therein to review the different components.
This facilitates model reporting by automating the generation of accompanying documents for all trained machine learning models and, in turn, expedites the review, audit, and approval of models, removing some of the burden typically associated with the sourcing and gathering of each of its underlying components.
For more information on EZOPS’ Model Cards or how our machine learning capabilities can boost your organization’s productivity contact email@example.com.