ML4CPS 2027: Machine Learning 4 Cyber Physical Systems Fraunhofer Forum, Anna-Louisa-Karsch-Straße 2 Berlin, Germany, March 11, 2027

Modern cyber-physical systems can adapt to evolving requirements, accommodate architectural changes throughout their lifecycle, and make sense of the heterogeneous data they generate. Combined with machine learning, this opens up powerful possibilities — from predictive maintenance and self-optimization to fault diagnosis, re-planning, and reconfiguration. Yet turning these ideas into well-founded, reliable methods remains an open challenge, requiring continuous research at the intersection of both fields. This workshop brings together researchers from the CPS and ML communities to discuss recent advances, open problems, and future directions — join the conversation.

The 10th Machine Learning for Cyber-Physical Systems (ML4CPS) workshop will take place on March 11, 2027 at the Fraunhofer Forum in Berlin. This year, ML4CPS will be held jointly with the 1st Machine Learning for Defence (ML4D) workshop, which follows on March 12, 2027.

ML4CPS is hosted by Fraunhofer IOSB, Helmut Schmidt University Hamburg, Hamburg University of Technology, and the Chair of Production Engineering of E-Mobility Components (PEM) at RWTH Aachen.

Submission Guidelines for the ML4CPS

All papers undergo a peer-review process. To be considered for presentation at the conference, please submit an extended abstract of up to two pages through the conference portal. Upon acceptance, authors are invited to submit a full paper (max. 15 pages) for publication in the conference proceedings by Helmut Schmidt University Press (openHSU), which will receive a unique DOI. Papers of a commercial nature will not be considered.

For additional details and submission guidelines for the ML4CPS and ML4D, please refer to: www.hsu-hh.de/imb/en/ml4cps

Papers may cover, but are not limited to the following topicsPapers may cover, but are not limited to the following topics

• Agentic AI & Multi-Agent Systems for CPS: Autonomous, tool-using agents that independently plan, diagnose, and act go beyond text- and multimodal-focused LLM-agents, opening new possibilities for intelligent, self-directed operation in cyber-physical systems.
• Time-Series Foundation Models: Foundation models specialized for time-series data enable new approaches to predictive maintenance, anomaly detection, and forecasting, addressing the unique challenges of sensor and process data.
• Industrial AI: Integrating AI into manufacturing processes can help to optimize them and enhance operational efficiency. Still, integrating AI into legacy systems and existing infrastructure is still a major challenge.
• Green AI: Reducing the energy consumption of AI systems is essential for industrial and edge applications. This topic focuses on methods for energy-efficient models, and the trade-off between performance and resource usage.
• Hybrid Methods & Hybrid Systems: Hybrid methods integrate multiple learning and modeling techniques while hybrid systems combine discrete and continuous dynamics and, thus, are powerful paradigms for complex CPS and industrial processes. Methods related to data-driven model identification, diagnosis, verification, and analysis are relevant challenges for the community.
• Simulation-to-Real / Synthetic Data: As real-world data for CPS is often scarce or costly, simulation-to-real transfer and synthetic data generation are key enablers for training and validating robust models, complementing physics-inspired approaches.

Contact

All questions related to paper submissions should be emailed to ml4cps_orga@hsu-hh.de