@masterthesis{Schulz2021, type = {Bachelor Thesis}, author = {Schulz, Theresa}, title = {Konzeption und prototypische Entwicklung eines Thesaurus f{\"u}r IT-Konzepte an Hochschulen Bachelorarbeit}, doi = {10.25933/opus4-2732}, url = {http://nbn-resolving.de/urn:nbn:de:kobv:522-opus4-27323}, school = {Technische Hochschule Brandenburg}, year = {2021}, language = {de} } @article{SchulzJander2025, author = {Schulz, Theresa and Jander, Kai}, title = {Dynamic plan generation with LLMs: automatic execution of abstract BDI-agent goals}, series = {International Journal of Parallel, Emergent and Distributed Systems}, journal = {International Journal of Parallel, Emergent and Distributed Systems}, publisher = {Taylor\&Francis}, doi = {10.1080/17445760.2025.2541956}, pages = {1 -- 21}, year = {2025}, abstract = {Belief-Desire-Intention (BDI) agents enable complex behavior in multi-agent systems by defining goals and implementing plans to achieve their goals. Typically, developers pre-define plans, which limits adaptability. This paper presents an approach for BDI agents to dynamically generate plans using large language models (LLMs), exploiting contextual information such as goals, beliefs, and structural data. We evaluate the feasibility, reliability, and limitations of this method, and discuss its implications for the automation of BDI-based agents. Our results suggest that LLMs can enhance agent autonomy by reducing the need for manual plan definition, while maintaining goal-oriented reasoning.}, language = {en} } @inproceedings{HelmichBraubachSchulzetal.2025, author = {Helmich, Roland and Braubach, Lars and Schulz, Theresa and Jander, Kai}, title = {Securing End-to-End Encrypted File Sharing Services with the Messaging Layer Security Protocol}, series = {Intelligent Distributed Computing XVII, 2024}, booktitle = {Intelligent Distributed Computing XVII, 2024}, number = {1203}, publisher = {Springer}, doi = {10.1007/978-3-031-87639-4_7}, pages = {107 -- 122}, year = {2025}, abstract = {To protect data on the servers of cloud service providers, file-sharing services rely on End-to-End Encryption (E2EE). However, existing solutions have weaknesses that allow attackers to bypass E2EE permanently after stealing a clients keys once. In this paper, a concept for an E2EE file-sharing service is proposed which does not have this vulnerability. It is based on Messaging Layer Security (MLS) groups for key distribution, an authentication system based on asymmetric cryptography, Attribute-Based Access Control (ABAC) based access rights and a tamper-proof versioned storage system for synchronising sensitive data. The applicability of the concept is demonstrated by a prototype implementation and an evaluation based on benchmarks and a security analysis. Overall, the concept can fulfil the requirements of a basic file sharing service while providing stronger security guarantees than existing solutions.}, language = {en} } @inproceedings{SchulzJander2025, author = {Schulz, Theresa and Jander, Kai}, title = {Planless Agents: BDI Plan Generation Using LLMs}, series = {Intelligent Distributed Computing XVII, 2024}, booktitle = {Intelligent Distributed Computing XVII, 2024}, number = {1203}, publisher = {Springer}, doi = {10.1007/978-3-031-87639-4_15}, pages = {253 -- 268}, year = {2025}, abstract = {Belief-Desire-Intention (BDI) agents offer a unique approach for engi- neering complex behavior for individual agents in multi-agent systems. A developer can define goals for each agent, specifying the desired outcomes in various con- texts and implement plans as the means to reach those goals. The BDI reasoning engine can then proceed to automatically select goals to pursue (goal deliberation) and choose one or more of the provided plans to attempt to achieve them. However, usually plans or at least part of the plans have to be provided by the agent's developer before the system is deployed. In this paper we present an approach for BDI agents to generate their own plans using large language models (LLMs) using solely the available context information such as goal descriptions, available beliefs and their structural information. We show that the approach is viable in principle and explore its reliability as well as discuss further use of LLMs in the context of automating BDI-based agents.}, language = {en} }