MLOps Enablement with Red Hat OpenShift AI (AI500)

Agenda

1 - What is MLOps?

• Brainstorm and explore what principles, practices, and cultural elements make up a MLOps model for ML model developments and deployments.

2 - Inner Loop

• Familiarize ourselves with the necessary tools for experimenting and building our model; we will create a workbench, explore the dataset, start tracking our experiments, and deploy our models.

3 - Training Pipelines

• Transition to automating the previous steps for productionizing our model training.

4 - Outer Loop

• Introduction to MLOps: a set of practices that automate and simplify machine learning workflows and deployments.
• Here we will create our MLOps environment where the continuous training pipeline, automated deployment, and the supporting toolings will be running.

5 – Monitoring

• Machine learning models can be influenced by various factors, including changes in data patterns, shifts in user behavior, and evolving external conditions. By implementing continuous monitoring, we will proactively identify these changes, assess their impact on model accuracy, and make necessary adjustments to maintain optimal performance.

6 - Data Versioning

• Enhance traceability by introducing versioning for our datasets as they change over time.

7 - Advanced Deployments

• Properly handle pre- and post-processing for data and predictions, explore autoscaling to handle loads, and introduce advanced deployment patterns like canary and blue-green deployments to ensure safe and seamless model rollouts.

8 - Feature Stores

• Robust ways of dealing with data features and their changes, as well as making sure features are homogeneous between training and serving.

9 - Security

• Implement automated security guardrails to stay compliant with the organizations security practices and extend them to the models.