Abstract of the presentation by Julian Rüpp from Bosch:

The rapidly evolving field of machine learning offers an overwhelming variety of modeling options. To make the selection process less cumbersome, a structured approach is required. This talk will explore the journey from initial modeling competition to the deployment of a relevant ML solution. We begin by emphasizing the importance of setting clear objectives and establishing fixed competition datasets, which serve as the foundation for consistent model evaluation. By logging models in a shared space, we ensure that all models are accessible by a uniform evaluation method, facilitating fair comparisons. Collaborating with clients to develop evaluation soft metrics is crucial in determining which model is best suited for scaling. As we move towards production, the focus shifts to enhancing robustness through orchestrated workflows, while also considering cost efficiency and inference time. Join us as we delve into the journey from competition to a final model, ensuring not only technical performance but also meet the practical demands of deployment.