Saltar al contenido principal

Model Governance

Updated May 14, 2023 ·

Overview​

Model governance provides proper oversight of ML models, especially when decisions have high consequences.

  • Manages access, policies, and activity tracking.
  • Reduces business risks and maintains reputation.

While it may slow down development, model governance ensures safe and valuable ML model use.

Design Phase​

In the design phase, we address questions like:

  • When is it ethical to use ML models?
  • How do we ensure privacy and control over sensitive data?
  • How do we avoid model bias?

Development Phase​

During development, we focus on:

  • Documenting the model selection process.
  • Ensuring data quality and versioning.
  • Making the model reproducible for testing and future use.

Pre-Production​

Before production, we must:

  • Prove the model's API is secure.
  • Set up monitoring and alerting systems.
  • Plan for failure handling and define who is responsible for each action.

This phase ensures that the model is secure and can be maintained once live.

Regulatory Considerations​

Some industries, like finance, require strict compliance with regulations to avoid penalties. The risk level of the model impacts how stringent the governance needs to be.

Higher-risk models, like those for detecting fraud, face more governance requirements than low-risk models, such as product recommendations.