Agentic AI for Complex Goals (T1)

Research Challenges / Novelty / Innovation
Challenges: learning from sparse feedback, reasoning in multi-domain contexts (engineering, medicine, logistics), robust agent cooperation: value alignment, conflict resolution, avoiding systemic errors

Research results:
• Development of a robust design framework for Agentic AI (Ontology of errors, design pattern and agent interaction library, unified LLM + knowledge graph integration pipeline.)

Innovate:
• AI Agentic Paradigm focused on complex goals, hierarchical planning, and self-adaptation.
• Combining Symbolic and Parametric AI paradigms (LLMs + ontologies/graphs).

Eye-on-AI: Explaining Automated and Semi-Automated Diagnostic Models in Medicine (T5)

Research Challenges / Novelty / Innovation
Understanding disease mechanisms requires integrating biomedical knowledge with symbolic and probabilistic AI.
• Major challenges in the explainability of AI decisions in medicine: lack of alignment with clinical reasoning, explanatory biases, heterogeneous data, lack of standardization.

Research results:
• Creation of a benchmark set for testing hypotheses and explanations
• Building an interactive XAI demonstrator/dashboard for doctors
• Publishing an open library with XAI methods adapted for the medical context
• 2 prototypes

Innovate:
• Unified approach „systems biology” applied to studies omics(in specialized proteomics), for clarification of contextual mechanisms in eye diseases.