Research Associate / Doctoral Student
Contact: doernfeld(at)peasec.tu-darmstadt.de | ORCID
Technical University of Darmstadt, Department of Computer Science,
Science and Technology for Peace and Security (PEASEC)
Pankratiusstraße 2, 64289 Darmstadt, Room 116
EN
Timon Dörnfeld, M.Sc. ist wissenschaftlicher Mitarbeiter und Doktorand am Lehrstuhl Wissenschaft und Technik für Frieden und Sicherheit (PEASEC) im Fachbereich Informatik der Technischen Universität Darmstadt. Seine Forschungsgebiet ist die Resilience von Kommunikationsinfrastruktur in vulnerablen sozio-technischen Systemen im Kontext von hybriden Bedrohungen. Darüber hinaus verfolgt er Fragestellungen zu Mensch-Computer-Interaktion, KI und Maschinellem Lernen.
Er studierte Physik an der Technischen Universität Darmstadt und der Königlich Technischen Hochschule in Stockholm im Frühjahr 2018. Im Nebenfach beschäftigte er sich mit dialektisch-materialistischer Philosophie, sowie Politikwissenschaften und Internationalen Beziehungen. Sein Abschluss erlangt er im Frühjahr 2021. Seine Masterarbeit befasste sich mit Symmetriebrechung in neutronenreicher Materie, wie sie z.B., im Inneren von Neutronensternen diskutiert wird.
DE
Timon Dörnfeld, M.Sc. is a research assistant and doctoral candidate at the Chair of Science and Technology for Peace and Security (PEASEC) in the Department of Computer Science at Darmstadt Technical University. His research focuses on the resilience of communication infrastructure in vulnerable socio-technical systems in the context of hybrid threats. He also pursues questions related to human-computer interaction, AI and machine learning.
He studied physics at Darmstadt Technical University and the Royal Institute of Technology in Stockholm in spring 2018. He minored in dialectical materialist philosophy, political science, and international relations. He graduated in spring 2021. His master’s thesis dealt with symmetry breaking in neutron-rich matter, as discussed, e.g., in the interior of neutron stars.
Publications
2025
[BibTeX] [Abstract] [Download PDF]
The network of subsea data cables (SDC) transmits the majority of international and intercontinental data exchanges. After thirty years of fiber-optic SDC installation across the oceans, almost all coastal and island countries gained access to the only global fixed infrastructure network. Still, there is considerable inequality in the number of available SDC accesses, creating deficits in redundancy for less connected states. Previous research hypothesized multiple factors that influenced the build-up of internet infrastructures but failed to verify these assumptions through inferential statistics. This work highlights the national-level factors that made backbone access provision more – or less – attractive to SDC project decision-makers. Our regression analysis of global country-year data (n = 4916) found that socio-economic (population, GDP), political (state fragility, conflict), and geographic factors (seismic hazard, neighboring territories) significantly influenced the number of active and planned accesses. This work can serve as a foundation for further research leveraging quantitative statistics to unveil hidden structures in the construction of material internet infrastructures and support sustainability in the future allocation of international infrastructure development resources in general.
@article{frankenHiddenStructuresGlobal2025,
title = {Hidden structures of a global infrastructure: {Expansion} factors of the subsea data cable network},
volume = {215},
issn = {0040-1625},
shorttitle = {Hidden structures of a global infrastructure},
url = {https://www.peasec.de/paper/2025/2025_FrankenReinholdDörnfeldReuter_TechForecasting.pdf},
doi = {10.1016/j.techfore.2025.124068},
abstract = {The network of subsea data cables (SDC) transmits the majority of international and intercontinental data exchanges. After thirty years of fiber-optic SDC installation across the oceans, almost all coastal and island countries gained access to the only global fixed infrastructure network. Still, there is considerable inequality in the number of available SDC accesses, creating deficits in redundancy for less connected states. Previous research hypothesized multiple factors that influenced the build-up of internet infrastructures but failed to verify these assumptions through inferential statistics. This work highlights the national-level factors that made backbone access provision more – or less – attractive to SDC project decision-makers. Our regression analysis of global country-year data (n = 4916) found that socio-economic (population, GDP), political (state fragility, conflict), and geographic factors (seismic hazard, neighboring territories) significantly influenced the number of active and planned accesses. This work can serve as a foundation for further research leveraging quantitative statistics to unveil hidden structures in the construction of material internet infrastructures and support sustainability in the future allocation of international infrastructure development resources in general.},
urldate = {2025-03-03},
journal = {Technological Forecasting and Social Change (TFSC)},
author = {Franken, Jonas and Reinhold, Thomas and Dörnfeld, Timon and Reuter, Christian},
month = jun,
year = {2025},
keywords = {Peace, A-Paper, AuswahlPeace, Security, Selected, Projekt-ATHENE-SecFOCI},
pages = {124068},
}[BibTeX] [Abstract] [Download PDF]
This paper investigates the potential of Large Language Models (LLMs) in automating the extraction and analysis of Subsea Data Cables (SDCs)-related information from unstructured media sources. A comprehensive LLM-based information extraction module was developed and integrated into an existing SDC database. By systematically comparing different LLMs, including GPT-4o, Gemini 1.5 Flash, Claude 3.5 Sonnet, and Llama 3.1, this work identifies optimal model configurations for SDC news processing, considering accuracy, hallucination rate, and processing speed. The system implements Claude 3.5 Sonnet and GPT4-o as primary models, incorporating domain-specific prompt engineering and robust output validation mechanisms. Performance evaluation demonstrates substantial improvements over rule-based methods. While excelling at natural language understanding tasks, the system revealed limitations in extracting more specific technical details such as construction costs and capacity measurements. The implementation provides a modular, adaptable framework for automated information extraction in specialised technical domains. The results demonstrate that LLMs, when properly implemented with structured prompts and validation mechanisms, can significantly enhance the automated monitoring and analysis of SDC-related events.
@inproceedings{frankenLinesLeveragingLLMs2025,
address = {Rostock, Germany},
title = {Between the {Lines}: {Leveraging} {LLMs} for {Information} {Extraction} of {Subsea} {Data} {Cable} {News}},
url = {https://zenodo.org/records/17119936},
doi = {10.5281/zenodo.17119935},
abstract = {This paper investigates the potential of Large Language Models (LLMs) in automating the extraction and analysis of Subsea Data Cables (SDCs)-related information from unstructured media sources. A comprehensive LLM-based information extraction module was developed and integrated into an existing SDC database. By systematically comparing different LLMs, including GPT-4o, Gemini 1.5 Flash, Claude 3.5 Sonnet, and Llama 3.1, this work identifies optimal model configurations for SDC news processing, considering accuracy, hallucination rate, and processing speed. The system implements Claude 3.5 Sonnet and GPT4-o as primary models, incorporating domain-specific prompt engineering and robust output validation mechanisms. Performance evaluation demonstrates substantial improvements over rule-based methods. While excelling at natural language understanding tasks, the system revealed limitations in extracting more specific technical details such as construction costs and capacity measurements. The implementation provides a modular, adaptable framework for automated information extraction in specialised technical domains. The results demonstrate that LLMs, when properly implemented with structured prompts and validation mechanisms, can significantly enhance the automated monitoring and analysis of SDC-related events.},
booktitle = {Proceedings of the {MARESEC} 2025},
author = {Franken, Jonas and Romer, Kasimir and Dörnfeld, Timon and Meissner, Paula and Reuter, Christian},
year = {2025},
keywords = {Projekt-emergenCITY, Student, Security, emergenCITY\_INF, emergenCITY\_SG, Projekt-ATHENE-SecFOCI},
}