Consolidating MLOps at Schiphol Airport Using MLFlow

Explore how Schiphol Airport leverages MLFlow to manage mission-critical machine learning models in production during a 50-minute conference talk. Discover the airport's approach to quickly iterating on models, from predicting passenger flow to analyzing aircraft surroundings using computer vision. Learn about the integration of MLFlow with other systems, including Airflow operators for model retraining and CI pipeline integration. Gain insights into the airport's controlled model release process, achieving software CI pipeline-like benefits and speed. Understand the motivation behind their MLOps strategy, training setup, inference running, and model monitoring techniques. Delve into topics such as feature compatibility, atomic deployments, and automated retraining. Uncover the key takeaways and benefits of consolidating MLOps in a large-scale airport environment.

Intro
Schiphol Airport
Motivation
MlFlow training set-up
There is no DEV in machine learning
Running Inference
Stability Assumption on your codebase
Feature compatibility
Atomic deployments
Model registry
Automated Retraining
Model monitoring
Benefits
Key takeaways