Toward Safeguarding GenAI with Graph Transformation: Defining an Exchange Format in JSON (Lightning Talk)
Generative Artificial Intelligence (GenAI), in particular Large Language Models (LLMs), have shown promising capabilities for various software engineering tasks, such as code completion, test gener ation, and Model-Driven Engineering (MDE). To improve the results, existing proprietary LLMs like GPT, Gemini, and Deepseek can be safeguarded with techniques such as PurpleLlama and NeMo-Guardrails. However, these techniques offer no guarantees for the syntactic or semantic correctness of complex structures such as graphs or, more specifically, Petri nets.
Our vision is a modular graph transformation-based approach for safeguarding LLMs, which guaran tees that the output (i) is syntactically and semantically correct, (ii) is complete so that a specified source graph can be transformed into a specific target graph using a set of generated graph transformation rules, and (iii) is traceable and explainable in multi prompt cases, where each part of an output can be linked to a certain (part of a) prompt. This is especially useful when complex graph structures like business workflows, class diagrams or module dependencies need to be generated, analysed, or changed, and the results must satisfy a set of user-specified constraints.
Tue 10 JunDisplayed time zone: Amsterdam, Berlin, Bern, Rome, Stockholm, Vienna change
15:30 - 17:00 | GCM Session 3: Lightning Talks & Panel Discussion: Graph Transformation and AIGCM at M 201 Chair(s): Reiko Heckel University of Leicester, Leen Lambers Brandenburg University of Technology Cottbus-Senftenberg, Oszkár Semeráth Budapest University of Technology and Economics Lightning talk: 5’ presentation and max. 5’ Q&A Subsequent panel & open discussion | ||
15:30 10mTalk | Pushing the boundary where formal language theory meets AI-assisted reasoning (Lightning Talk) GCM Federico Vastarini University of Salerno | ||
15:40 10mTalk | Toward Safeguarding GenAI with Graph Transformation: Defining an Exchange Format in JSON (Lightning Talk) GCM Lukas Sebastian Hofmann Philipps-Universität Marburg | Universidad Complutense de Madrid, Alexander Lauer Philipps-Universität Marburg, Jose Ignacio Requeno Complutense University of Madrid, Gabriele Taentzer Philipps-Universität Marburg | ||
15:50 10mTalk | Towards Graph-Based Neuro-Symbolic Logic Reasoning to Improve AI Applications (Lightning Talk) GCM Kristóf Marussy Budapest University of Technology and Economics | ||
16:00 60mPanel | Panel & open discussion: Graph Transformation and AI GCM Kristóf Marussy Budapest University of Technology and Economics, Gabriele Taentzer Philipps-Universität Marburg, Federico Vastarini University of Salerno |