The KI2026 conference took place in Bremen from 11–14 August 2026, bringing together researchers, developers, and decision-makers from academia, industry, and public administration to discuss current developments in artificial intelligence.
Novruz Mammadli from the University of Oldenburg’s Chair of Applied Artificial Intelligence and the Interactive Machine Learning group, presented “Multi-species Mixing for Weakly Supervised SED Under Domain Shift”. The paper addresses the challenge of making AI models for acoustic biodiversity monitoring more robust under real-world conditions. By synthetically mixing recordings of different frog species during training, the researchers improved model transfer to real monitoring data without requiring additional annotations. The results showed gains in both recall and F1-score, highlighting the benefits of simulating realistic multi-species scenarios during training.
Novruz Mammadli at the KI2026 in Bremen
Pratik Sitapara also presented “Grounded Label Space Engineering for Knowledge-Centric Annotation Workflows” at ARDUOUS 2026, an international workshop on annotation of real-world data for AI systems, held as part of KI2026. The paper introduces Grounded Label Space Engineering (GLSE) as a methodological framework that connects data annotation with knowledge engineering. The work addresses limitations of conventional annotation workflows, where label spaces are often flat, fixed, and weakly grounded in shared semantic resources. GLSE treats label spaces as evolving, ontology-linked semantic objects, supporting the elicitation and refinement of concepts during annotation, grounding them in authoritative knowledge resources where possible, and representing genuinely novel concepts as modular ontology representations with provenance under expert supervision.
Together, the contributions highlight how robust AI methods and structured knowledge can support more adaptive and reliable AI systems for real-world applications.
Pratik Sitapara presented his research at ARDUOUS @ KI2026
