MLOps: Keeping models from misbehaving like angry teenagers By Cedric De Haes, Bert Gossey

As the field of machine learning continues to skyrocket and AI has become the talk of the town, it becomes clear that the way models are being developed, verified, and put into production needs to grow up – literally. Just like how the (angry) teenager slowly reaches adulthood, we’re drawing ever closer to a robust, mature model development lifecycle to cope with the continuous retraining of models and the fast-moving changes in the field. Luckily, we can draw from the wisdom of software developers and their techniques to get us there. In this talk, Bert and Cédric will explore the wild world of MLOps – what it entails, how it is similar to, but also different from software engineering practices. We’ll show you the immense, potential benefits of having them work together with data science and machine learning teams, all from an app developer’s standpoint. By examining the similarities and differences between these disciplines, you can expect to go home carrying a deeper understanding of the opportunities presented by MLOps. Together, we’ll explore: · model version control · CI/CD · data pipelines · monitoring tools · scalable and reliable deploying (think kubernetes) And the best part? You don't need (much) machine learning expertise to join us for this talk. See you at Devoxx? CEDRIC DE HAES Coming from the world of data science, Cédric De Haes is a software engineer with a keen interest in exploring the intersection of machine learning and software development. In his free time, it’s hard to keep Cédric from experimenting with models, further deepening his knowledge and expertise. BERT GOSSEY For the past 20 years, Bert has been working in the technology realm, with the last few years working on (big) data and AI projects. He relies on his scientific background to blend the worlds of software engineering and machine learning into his own signature brand of innovation.