Green AutoML for Driver Assistance

Nowadays, AI applications can be found in many devices used in daily life, which means that the average energy consumption of a person is constantly increasing. Due to the scarcity of resources, it is thus increasingly important to also develop AI applications in a resource-saving manner. However, in this context, a major challenge includes analyzing large amounts of data with security relevance. Due to its complexity, deep learning, frequently used for this purpose, usually requires high energy consumption and thus generates a large ecological footprint. In order to prevent this footprint from becoming too large, resource-efficient AI applications are absolutely necessary. As an example, in our project, we study driver assistance systems, which improve the safety, comfort, and economy of driving.

The aim of the GreenAutoML4FAS project is to design a holistic system consisting of hardware, efficient coding and transmission of data and models, and dynamic and adaptive software in a resource-efficient manner. To this end, we will develop new resource-efficient AutoML systems that efficiently support developers in the entire AI development cycle. Exemplarily, the focus here is on driver assistance systems. Combining efficient algorithms, communication, and hardware in this area will lead to significant energy savings. Thus, the holistic concept developed in the project will also be transferred to other areas in which AI or deep learning is used as a machine learning method.

Lead at LUHAI: Prof. Lindauer

Funding Program: AI Beacons for the Environment, Nature and Resources, Bundesministerium für Umwelt, Naturschutz, nukleare Sicherheit und Verbraucherschutz

Funding Reference: 67KI32007A

Project Period:  March 2023 to February 2026

Project Partners:

• Institute of Artificial Intelligence (Prof. Dr. Marius Lindauer)
• Institute for Information Processing (Prof. Dr. Bodo Rosenhahn, Prof. Dr. Jörn Ostermann)
• Institute of Microelectronic Systems (Prof. Dr. Holger Blume)
• VISCODA GmbH (Dr. Hellward Broszio)

Project Executing Agency: ZUG

• Neutatz, Lindauer, Abedjan. "[Experiments \& Analysis] How Green is AutoML for Tabular Data?". Proceedings of EDBT 2025

• Giovanelli, Tornede, Tornede, Lindauer. "Interactive Hyperparameter Optimization in Multi-Objective Problems via Preference Learning". Proceedings of the AAAI Conference on Artificial Intelligence (Vol. 38. No. 11. 2024).
• Hennig, Tornede, Lindauer. "Towards Leveraging AutoML for Sustainable Deep Learning: A Multi-Objective HPO Approach on Deep Shift Neural Networks". In PML4LRS Workshop @ ICLR (2024).
• Norrenbrock, Rudolph, Rosenhahn. “Q-SENN: Quantized Self-Explaining Neural Networks”. In Proceedings of the AAAI Conference on Artificial Intelligence (Vol. 38, No. 19, 2024).
• Theodorakoloulos, Stahl, Lindauer. "Hyperparameter Importance Analysis for Multi-Objective AutoML". In Proceedings of ECAI (2024).
• Tornede, Deng, Eimer, Giovanelli, Mohan, Ruhkopf, Segel, Theodorakopoulos, Tornede, Wachsmuth, Lindauer. “AutoML in the Age of Large Language Models: Current Challenges, Future Opportunities and Risks”. In Transactions on Machine Learning Research (2024).
• Neutatz, Lindauer, Abedjan. "AutoML in Heavily Constrained Applications". VLDB Journal (2024).

• Brockmann, Rudolph, Rosenhahn, Wandt. “The Voraus-AD Dataset for Anomaly Detection in Robot Applications”. Transactions on Robotics (2023).
• Glandorf, Kaiser, and Rosenhahn. “HyperSparse Neural Networks: Shifting Exploration to Exploitation through Adaptive Regularization”. Proceedings of the IEEE/CVF International Conference on Computer Vision (2023).
• Loni, Mohan, Asadi, Lindauer. "Learning Activation Functions for Sparse Neural Networks". AutoML Conf (2023).
• Rosenhahn. “Optimization of Sparsity-Constrained Neural Networks as a Mixed Integer Linear Program: NN2MILP”. Journal of Optimization Theory and Applications (2023)