Setareh Maghsudi and her research team have completed their research project “Communication-constrained Cooperation under Uncertainty: How to Make Strategic Decisions?” The project was funded by the Cyber Valley Research Fund and was carried out between 2022 and 2025 at the University of Tübingen. The project’s aim was to create learning methods that are both scientifically sound and useful in real-world situations. These methods should also enable groups of independent systems to make good decisions in environments that are constantly changing, uncertain, and where communication is limited.
Many of today’s most important technologies, such as power grids, recommendation engines, autonomous technologies, and online marketplaces, consist of a large network of independently operating software systems. These systems need to learn, coordinate, and allocate resources, often without complete information and under various practical constraints.
To address this problem, Maghsudi and her research team developed new approaches that improve how distributed systems learn and cooperate. One key result is CLUB-HG, an algorithm that allows recommendation systems to continuously and automatically discover groups of users with similar preferences as new information becomes available. Rather than treating every user separately, the method allows those with similar preferences to learn from each other’s experiences. This enables large systems to scale more efficiently, because knowledge gained in one part of the system can benefit others facing similar situations. In applications such as recommendation platforms, this can help systems learn user preferences faster and deliver better recommendations with less data.
Another result was MA-LfL, a method that learns what different participants in a system are trying to achieve by observing their behavior. Unlike previous approaches, it can do this even when participants are still learning themselves rather than already behaving as perfect experts. This idea is reflected in the name MA-LfL, which stands for “Learning from Learners.” MA-LfL makes it possible to better understand the goals of different decisionmakers and how they interact, helping to build systems that cooperate more effectively even while their behavior is still changing.
The project’s most immediate practical impact lies in helping complex systems make better use of limited resources under uncertainty. This is particularly relevant for smart energy systems, where fluctuating renewable energy sources require constant coordination of limited resources. By improving decision-making under uncertainty, the methods could help increase efficiency, reduce waste, and lower operating costs.
This project produced the following peer-reviewed publications:
Cheng, X., Pan, C., and Maghsudi, S. (2023). Parallel Online Clustering of Bandits via Hedonic Game. International Conference on Machine Learning (ICML).
Kaliscan, M., Chini, F., and Maghsudi, S. (2023). Multi-Agent Learning from Learners. International Conference on Machine Learning (ICML).
Barghi, H., Cheng, X., and Maghsudi, S. (2023). Cooperative Thresholded Lasso for Sparse Linear Bandit. European Conference on Artificial Intelligence (ECAI).
Tsetis, I., Cheng, X., and Maghsudi, S. (2023). A Bandit Online Convex Optimiyation Approach to Distributed Energy Management in Networked Systems. IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP).
Ehyaei, AR., Karimi, AH., Schölkopf, B., and Maghsudi, S. (2024). Robustness Implies Fairness in Causal Algorithmic Recourse. ACM Conference on Fairness, Accountability, and Transparency (FAccT).
Cheng, X., Tsetis, I., and Maghsudi, S. (2025). Distributed Management of Fluctuating Energy Resources in Dynamic Networked Systems. IEEE Transactions on Network Science and Engineering.
About the Cyber Valley Research Fund
The Cyber Valley Research Fund was established to support Cyber Valley research groups undertake basic research in the fields of artificial intelligence and robotics. The fund totaled five million euros, including contributions from six of Cyber Valley’s founding corporate partners: Amazon, BMW, Bosch, IAV, Mercedes-Benz, Porsche, and ZF. It supported 20 research projects, the first of which began in 2020, and the final of which will conclude in 2026.
Note on AI use: Cyber Valley GmbH developed this document using generative AI as a supporting tool. Cyber Valley GmbH is responsible for its content and conclusions.


