Protecting Your Machine Learning Against Drift - An Introduction

Explore the critical topic of protecting machine learning models against drift in this 49-minute EuroPython Conference talk. Gain a practical introduction to drift detection, understanding how it occurs, its importance, and methods for principled detection. Delve into the challenges of detecting drift in high-dimensional, unlabelled data arriving continuously in deployment settings. Learn how to apply theoretical concepts using the alibi-detect Python library through a practical demonstration. Suitable for those with basic machine learning knowledge, covering topics such as statistical hypothesis testing, offline and online drift detection, windowing strategies, and test statistics.