Abstraction techniques for scalable verification of deep neural networks in Ingolstadt bei AUDI AG

The Dependable AI Systems team at AUDI AG is pursuing cutting edge multidisciplinary research in dependability analysis and safety verification of learning-based AI systems with a particular focus on providing methods and tools for the development of Audi next-generation highly automated driving. Your involvement in the pre-development of automated driving at AUDI AG enables you to gain experience in this rapidly growing field and to participate in the development of conditionally and highly automated driving functions in an environment that is both scientifically oriented and practical at the same time.

The increasing use of deep neural networks (DNNs) as controls for safety-critical systems such as autonomous driving has posed the challenge of verifying their correctness. The latter is mainly due to the opacity of DNNs. Several verification approaches have been proposed to address this problem. However, due to the problem of scalability, their direct application in large networks is limited. To take the latter into account and as part of your internship, you will develop new abstraction techniques to improve the scalability of verification algorithms.

Job Purpose/Role

• Learning and mastering the relevant state of the art
• Concept development and prototypical implementation of algorithms
• Evaluation of the methods based on measured data
  
  The exact topic is defined in collaboration with the supervising university.
  

Key Requirements/Skills/Experience

• Course of studies in a technical area (e. g. computer science, mathematics or comparable)
• Excellent academic track record
• Domain knowledge in deep learning or verification
• Demonstrated experience in at least one of the following topics: machine learning, neural networks; computer vision; optimization theory; control theory or formal verification
• Very good programming skills in Python or C++
• Experience in AI and big data frameworks such as TensorFlow, Keras, Scikit-learn, Spark ML
• The ability to work autonomously in an international environment, creativity, team spirit and a pro-active attitude are imperative