Standort Ottobrunn bei München Beschreibung der Stelle The IABG Innovation Centre serves as a development incubator for IABG’s portfolio, focusing on key trends such as digitalisation, artificial intelligence, robotics, and sensor networks in the mobility and security domains.
As part of the safeAI program, the Innovation Centre explores how cutting-edge AI algorithms can be aligned with current and emerging AI assessment frameworks. Given the rapid pace of advancement in this field, new AI frameworks, software libraries, tools, and updates are continuously released. These must be systematically reviewed, evaluated, and integrated into our workflows where appropriate.
Standardisation also plays a critical role in shaping the future of AI assurance services. IABG is actively contributing to this domain and is leading the development of ISO/IEC TS 25223, which provides guidance and requirements for the uncertainty quantification of AI systems.
This internship will focus on the research and implementation of uncertainty quantification methods for generative AI applications. The developed approaches will be evaluated through a use case in image recognition. The overarching goal is to deploy a vision-language model and demonstrate how uncertainty quantification techniques can enhance the reliability of its outputs for the selected task. Tätigkeit

• You review state-of-the-art vision-language models.
• You review state-of-the-art methods for uncertainty quantification in generative AI.
• You implement a fully functional pipeline for a specified use case.
• You implement selected uncertainty quantification method(s) for a vision-language model.
• You perform experiments and an initial evaluation for a specified use case.

Voraussetzungen

• You are currently pursuing a degree in computer science, data science, mathematics, or related engineering disciplines.
• You have strong programming skills in Python.
• You have good knowledge of Linux, Docker, CUDA, and Git.
• You have knowledge and experience in machine learning approaches and the foundations of data science.
• Experience with vision-language models and uncertainty quantification is a plus.