Data-Driven AV Development: Data Management and MLOps
, Senior SW Developer for Autonomous Driving, BMW AG
Machine learning enables self-driving cars to handle the infinite combinations of situations that can’t be covered by a set of rules. However, machine learning development entails much more than crafting a model architecture and learning its hyperparameters from examples. Data, the source of the examples, is the critical ingredient. We'll present BMW’s approach to designing and implementing a data management system and performing MLOps for computer vision in the car. We'll highlight the design considerations and cover key lessons learned from implementing the system, including fast training on an NVIDIA GPU cluster running on Kubernetes.