Multimodal Reasoning Biomedical Safety Master Thesis

Master Thesis:
Multimodal Knowledge and Reasoning for Biomedical Safety Applications

Safe Intelligence — this forms the core brandof the Fraunhofer Institute for Cognitive Systems IKS
. Connected cognitive systemsdrive innovation in many sectors, includingmobility, healthcare, and automation in industry.

Disruptive technologies such as artificial intelligence and quantum computing play a key rolehere. Fraunhofer IKS is conducting research toensure that these applications are reliable andverifiably safe. We consider resilience and intelligence to be part of the same process.

Be part of change

In this master thesis, you will work at the intersection of multimodal machine learning,knowledge representation, and Reasoning inbiomedical safety - contributing to a researchsystem that must reason reliably overheterogeneous data sources.

Your work will span three interconnected challenges:

• Multimodal data integration: combiningstructured data (graphs, ontologies, relational databases), unstructured text (scientific literature, clinical reports), and molecular ornumerical features into a unified reasoningframework.
• Knowledge-grounded inference: designingretrieval and reasoning pipelines that groundmodel output in interpretable, traceable knowledgepaths.
• Uncertainty quantification and faithfulness:developing methods to certify when modelpredictions are well-supported by evidence and communicating confidence in a way that supportshuman oversight.

You will engage with the full research lifecycle,literature review, problem formalization, system design, implementation, and empirical evaluationagainst published baselines. You will have the opportunity to contribute to an applied researchproject with real-world deployment context.

What you contribute

Essential

• Strong Python programming skills and goodsoftware engineering practices (modular design,version control, documentation)
• Solid foundation is one of the following:
  • Multimodal learning: experience fusingheterogeneous input types such as text, graphs,structured tables, or molecularrepresentations
  • Knowledge graphs: construction, graph data models, traversal, or graph databases (Neo4j /Cypher ideally)
  • Retrieval-augmented generation (RAG) andLLM integration (Lang Chain, Llama Index, or equivalent)
• Ability to independently read, understand,and synthesize primary research literature
• Structured, self-driven working style with attention to reproducibility

Advantageous

• Graph machine learning: graph neural networks(GNN / GAT / RGCN) or knowledge-graphembeddings
• Generative AI knowledge: agentic workflows and multi-agent systems
• Biomedical or life science domain knowledge(ontologies, clinical data formats, omicspipelines)
• Working with messy real-world data: missingvalues, label noise, domain shift

Profile

• Enrolled at a German university;
  The candidate should be able to come to the office in Garching at least one day a week.
• Especially suitable for M.Sc. Computer Science, Bioinformatics, Data Engineering,Computational Life Sciences, or relateddisciplines
• Genuine curiosity about trustworthy AI andits application in high-stakes domains

What we offer

• Approachable supervisors and integrationinto a dynamic interdisciplinary team spanning AIsafety, data science, and application domains.
• Hybrid set up with workplace at ourmodern institute building in Garching-Forschungszentrum, close to TU Munich.
• Hands-on experience with real-world data pipelines, HPC infrastructure, and Fraunhofer’sapplied research practices.
• Work on open scientific questions withinan active applied research project, with a clearpath toward publication at international venues.