Complementary research topics

Consideration of uncertainties in E/E architectures
Cooperative automated driving must consider environmental, perception, and vehicle uncertainties across functional, logical, software, and technical architecture levels. The research in this part of Beyond Validation AI focuses on designing resilient architectures, ensuring safety and operability throughout DevOps. Key questions address minimizing uncertainties in automotive E/E architectures.

Management of simulation and real-world data
This part of Beyond Validation AI addresses the lack of efficient similarity search in Object Storage Systems by leveraging vector databases and learned index structures. It explores machine learning-based optimization of vector indexes for static and dynamic datasets, including multimodal and incomplete data. The goal is a robust and efficient similarity search.

Resource-efficient implementation through smaller models
This part of Beyond Validation AI focuses on achieving smaller, efficient AI models for inference while maintaining prediction quality. The focus lies on intra-model optimization and inter-model optimization, including techniques such as knowledge distillation. The research investigates how both approaches improve resource efficiency on automotive hardware.

Hardware-oriented implementation of AI models
Deploying neural networks in cooperative automated driving environments requires meeting strict resource and timing constraints. This part of Beyond Validation AI explores hardware-aware optimizations across different platforms and compiler frameworks. It investigates automatic synthesis of algorithms and data layouts, as well as the impact of quantization on uncertainty quantification.