business triangle technical hexagon

From trained models to machine learning systems

Technical talk | English

Theatre 25: Track 1

Wednesday - 17.05 to 17.45 - Technical


Machine learning provides an amazingly effective family of techniques for solving real business problems. But solving problems with machine learning is only a small part of the challenge: to take advantage of ML, we need to incorporate predictive models and learning pipelines into real systems. This talk will introduce some of the challenges involved in going from a successful experiment to a maintainable system and show you how to confidently navigate safely from a notebook to a robust production environment. You’ll learn effortless techniques for publishing models and pipelines as microservices, some of the surprising and silent ways that machine learning systems can fail in production, and how you can take advantage of contemporary cloud infrastructure to automatically warn you as soon as things start to go wrong. You’ll leave with a better understanding of how to put machine learning systems in production, a roadmap of pitfalls to avoid, and actionable suggestions for implementing these techniques using Kubernetes and open-source tools.