Monitoring, re-training and replacing MLOps applications

MLOps for Business

Arne Jonas Warnke

Head of Emerging Curriculum

Operating machine learning models

$$ The steps of the operations phase

MLOps for Business

Pre-deployment

$$ Pre-deployment phase

MLOps for Business

Pre-deployment

  • Ensure models producing the same results everywhere
    • Includes containerization
  • Additional checks
    • Security

$$

Usually the task of the

  • Software engineer or machine learning engineer

$$

A picture of containers in a harbor

MLOps for Business

Deploying the MLOps application

$$ Deployment phase

MLOps for Business

Deploying the MLOps application

Deployment (via CI/CD)

  • Is largely automated
  • Consists of different (automated) tests
  • Allows to quickly reverse changes

$$

Again usually the task of the

  • Software engineer or machine learning engineer

$$

A women getting her food delivery

MLOps for Business

Monitoring the MLOps application

Monitoring phase

MLOps for Business

Monitoring the MLOps application

Model is live now and

  • Generates business value
  • Downtimes and issues are costly
  • Needs to be closely monitored
    • Largely automated

If the model fails

  • Different people might need to get involved

Be aware model quality deteriorates over time.

$$

A futuristic monitoring room

MLOps for Business

Re-training the machine learning models

Re-training phase

MLOps for Business

Re-training the machine learning models

Model performance usually deteriorates

  • Economic changes
  • Clients have different preferences

This requires us to re-train the models regularly

  • Can be partly automated

Main task of

  • Data scientist / machine learning engineer

$$

An image of a circle

MLOps for Business

Let's practice

MLOps for Business

Preparing Video For Download...