Testing ML Pipelines: How to Verify Data and Model Quality at Every Stage

ML models can make mistakes—and it is not always their fault. Data degrades, pipelines fail, and businesses lose money. How can this be prevented?

In this talk, I will explain how to test an ML pipeline at every stage, from data preparation to monitoring model performance in production.

We will explore key methods for assessing data and inference quality, including data quality checks and model degradation control.

You will learn how to enhance the stability of ML systems, reduce losses caused by data errors, and discover new opportunities in the field of testing.

All insights are based on real-world testing experience and industry best practices.