Introduction

Instruction and application

In Progress

Introduction

Imagine needing to investigate a decision that a machine learning (ML) model made last year, only to find no logs, no version records and no way to recreate its predictions.

In modern MLOps, the lifecycle of an ML model doesn’t end at deployment. A well-run lifecycle involves continuous oversight, structured change tracking, and eventual decommissioning. This unit will equip you with the skills to manage the evolution and retirement of your ML models, ensuring traceability, compliance, and responsible resource management.

Why does this unit matter?

ML models are not "set and forget" assets. Over time, their utility declines, regulatory requirements change, and better alternatives emerge. Maintaining a model that is no longer accurate or necessary is not only a waste of compute resources but also a source of operational and compliance risk.

Mastering change management and decommissioning allows you to maintain a clean, efficient, and auditable model estate. It ensures that every model in production is serving a purpose and that, when its time is up, it can be retired safely and systematically.

Learning objectives

By the end of this unit, you will be able to:

• Design and implement change management processes for updating and redeploying ML models.
• Set up comprehensive logging and monitoring systems to track ML model changes and version history.
• Develop model decommissioning protocols for safely retiring and archiving obsolete ML models.
• Apply ethical and regulatory considerations to model lifecycle management and archiving.

Action item: Pause and think

Think about the models currently in use at your organisation. Do you know which version is running? Is there a plan for when they should be replaced?

Questions & reflections

What are the potential risks of keeping an obsolete ML model running in production?

Type your reflection here...