11:30 - 12:15
We have built a training pipeline and deployed our first machine learning model into production. But our data science project is far from over. Even a production model faces many challenges. Unlike "traditional" software, the quality of a machine learning system deteriorates over time. A model that is deployed in production and not retrained will degrade. It will never work as well as it did on day one. Therefore, we need to monitor it and decide when to build and release a new version.
You’ll leave this talk with an understanding of how we monitor machine learning models in production. We will talk about fast deployment cycles, DevOps, and deciding if our model still delivers business value. You’ll learn about failure modes of ML models and how we can detect them. Also, we will explore the use of A/B-testing for machine learning and little deployment strategies that can prevent big disasters when you retrain your model.