Archived News

We congratulate Dr. Tayyaba Zainab on successfully defending her PhD thesis at Kiel University: "From Passive Data Collection to Sensor-Level Intelligence: TinyML in Constrained IoT Environments". This thesis was in part funded by MarData and conducted in collaboration with Geomar.  Supervision: Prof. Dr. Olaf Landsiedel, TU Hamburg & Kiel University Dr. Jens Karstens, Geomar Dr. Laura Harms, Kiel University  Reviewers: Prof. Dr. Olaf Landsiedel, Kiel University Prof. Dr. Anna Förster, Bremen University  Examiner: Prof. Dr.-Ing. Kevin Köser, Kiel University Chairman of the defense: Prof. Dr. Sören Pirk, Kiel University

"From Raw Waveforms to on-device Earthquake Detection: Real-Time Seismic Data Analysis for MCUs" was just accepted at ACM ACM Transactions on Internet of Things (ACM TIOT). The paper presents an on-device earthquake detection system for microcontrollers, showing how raw seismological waveforms can be analyzed using AI  in real time directly on resource-constrained MCUs instead of relying on cloud processing. It targets fast, practical seismic event detection at the edge, reducing latency, communication overhead, and infrastructure demands for earthquake monitoring deployments. Congratulations to the team: Tayyaba Zainab, Patrick Rathje, Laura Harms, Lukas Schattenhofer, Jens Karstens, Olaf Landsiedel.

"ThinkingViT: Matryoshka Thinking Vision Transformer for Elastic Inference" has just been accepted to CVPR! ThinkingViT brings thinking-style adaptive computation to Vision Transformers by letting the model “think more” only when an image is hard, and exit early when it is confident. Concretely, ThinkingViT activates a small subset of the most important attention heads for an initial prediction, and progressively expands to larger subsets only if the confidence (entropy) suggests that more computation is needed. To make each rethink more powerful, we introduce Token Recycling, which fuses previous-stage embeddings back into the input so the model refines its decision instead of starting from scratch. Congratulations and thanks to the team: Ali Hojjat, Janek Haberer, Soren Pirk, Olaf Landsiedel Preprint: https://lnkd.in/eGriuYXp GitHub: https://lnkd.in/eDd84SSc

Drone Ballet ("Drohnenballet") is a joint project by the company Zouber from Kiel and Hamburg University of Technology, funded by the Federal Ministry of Research, Technology, and Space (BMFTR), Germany, as part of UAM-Inno Region SH. We plan to develop a new generation of creative, efficiently plannable drone shows. It focuses on AI-based flight-path planning and dynamic swarm control so that shows can be generated automatically, interactively adapted during performances, and enriched with effects such as morphing between shapes and interactive elements. The technology is also intended as a building block for further urban air mobility applications, for example, in sea rescue, agriculture, or inspection, and offers an emission-free alternative to fireworks.

Yesterday (19.01.2026), German broadcaster ARD reported about the Geomar research cruise M215 (MULTI-MAREX Cruise 3) to the volcanoes around Santorini and our work on earthquake detection: As part of this cruise, our researchers successfully tested our embedded AI models for earthquake detection and analysis.  Tagesschau (from minute 9) Europamagazin (longer report) Tagesschau24 (longer report) This work builds, in part, on our joint publications SeismicSense and LightEQ with Geomar and was, in part, funded by MarData.