Dr.-Ing. Zeyd Boukhers

Lead of FAIR Data & Distributed Analytics Group

Fraunhofer Institute for Applied Information Technology (FIT)

Institute for Medical Informatics, University Hospital of Cologne

I believe that AI and data science should be both powerful and trustworthy. My research develops machine learning, semantic web technologies, knowledge management, and FAIR data infrastructure that work reliably across domains—from healthcare and biodiversity to scholarly data and cultural heritage. With dual appointments at Fraunhofer FIT and University Hospital of Cologne, I build systems that enable researchers in any field to harness AI, semantic interoperability, and data science responsibly, transparently, and effectively.

Publications Research Projects Research Group Contact

Research Interests

Leading the FAIR Data & Distributed Analytics Group at Fraunhofer FIT

FAIR Data

Data management, FAIR principles, data spaces

Machine Learning

Deep learning, neural networks, AI applications

NLP

Natural language processing, text mining

Data Sovereignty

Data governance, privacy, security

Research Topics

Explore my research areas - click to see related publications

Recent News

December 01, 2025

AI-BioDynamics Project Kicks Off!

Our German-Korean AI-BioDynamics project has officially launched, supported by BMBF and MSIT with a total budget of €1.3 million to build an intelligent platform for next-generation biodiversity monitoring. The consortium will work together through November 2028.

Learn more

View all news

Recent Publications

1. 2026 • Conference Paper

   Autonomous FAIR Digital Objects: From Passive Assertions to Active Knowledge

   Zeyd Boukhers , Oya Beyan , Cong Yang , Christoph Lange

   The 25th International Semantic Web Conference (ISWC 2026)

   PDF

   Boukhers, Z., Beyan, O., Yang, C., & Lange, C. (2026). Autonomous FAIR Digital Objects: From Passive Assertions to Active Knowledge. The 25th International Semantic Web Conference (ISWC 2026). https://arxiv.org/abs/2605.10370

   Boukhers, Zeyd, , and Oya Beyan, , and Cong Yang, , and Christoph Lange. "Autonomous FAIR Digital Objects: From Passive Assertions to Active Knowledge." The 25th International Semantic Web Conference (ISWC 2026), 2026.

   Boukhers, Zeyd, , and Oya Beyan, , and Cong Yang, , and Christoph Lange. "Autonomous FAIR Digital Objects: From Passive Assertions to Active Knowledge." The 25th International Semantic Web Conference (ISWC 2026) (2026).

   @inproceedings{Boukhers2026autonomous,
     title = {Autonomous FAIR Digital Objects: From Passive Assertions to Active Knowledge},
     author = {Zeyd Boukhers and
              Oya Beyan and
              Cong Yang and
              Christoph Lange},
     booktitle = {The 25th International Semantic Web Conference (ISWC 2026)},
     year = {2026},
     url = {https://arxiv.org/abs/2605.10370},
     publisher = {Springer},
     
   }

   Scientific knowledge published on the Web generally remains passive and depends on external infrastructure for validation, conflict resolution, and confidence updates. This paper introduces Autonomous FAIR Digital Objects, extending FAIR Digital Objects with portable policy execution, event announcements, and distributed agreement mechanisms based on Semantic Web standards including RDF-star, PROV-O, SHACL, ODRL, and ActivityStreams. Evaluation on thousands of FAIR Digital Objects derived from rare-disease knowledge sources examines conflict resolution, distributed execution, and robustness under adversarial conditions.

2. 2026 • Conference Paper

   Evolutionary Soups: Evolving Mixture-of-Experts for Multi-Objective LLM Alignment

   Lingxiao Kong , Steffen Staab , Cong Yang , Oya Beyan , Zeyd Boukhers

   Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing (EMNLP 2026)

   PDF

   Kong, L., Staab, S., Yang, C., Beyan, O., & Boukhers, Z. (2026). Evolutionary Soups: Evolving Mixture-of-Experts for Multi-Objective LLM Alignment. Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing (EMNLP 2026). https://arxiv.org/abs/2608.29978

   Kong, Lingxiao, , and Steffen Staab, , and Cong Yang, , and Oya Beyan, , and Zeyd Boukhers. "Evolutionary Soups: Evolving Mixture-of-Experts for Multi-Objective LLM Alignment." Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing (EMNLP 2026), 2026.

   Kong, Lingxiao, , and Steffen Staab, , and Cong Yang, , and Oya Beyan, , and Zeyd Boukhers. "Evolutionary Soups: Evolving Mixture-of-Experts for Multi-Objective LLM Alignment." Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing (EMNLP 2026) (2026).

   @inproceedings{Kong2026evolutionarysoups,
     title = {Evolutionary Soups: Evolving Mixture-of-Experts for Multi-Objective LLM Alignment},
     author = {Lingxiao Kong and
              Steffen Staab and
              Cong Yang and
              Oya Beyan and
              Zeyd Boukhers},
     booktitle = {Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing (EMNLP 2026)},
     year = {2026},
     url = {https://arxiv.org/abs/2608.29978},
     publisher = {Association for Computational Linguistics},
     
   }

   Large language models increasingly need to satisfy multiple competing objectives whose preferred trade-offs vary across users and prompts. Evolutionary Soups introduces a mixture-of-experts framework for controllable multi-objective generation in which per-layer gating networks dynamically determine expert-merging coefficients from hidden representations. The gating networks are optimized using an evolutionary algorithm with greedy hypervolume contribution. Across three tasks, the method improves Pareto-front coverage and achieves strong hypervolume, linear-utility, and Tchebyshev-utility results compared with existing controllable approaches.

3. 2026 • Conference Paper

   Automated Construction of FAIR Digital Object Knowledge Graphs from Flat Cultural Heritage Records

   Zeyd Boukhers , Lingxiao Kong , Xenophon Zabulis , Georgios Toubekis

   Proceedings of the 35th ACM International Conference on Information and Knowledge Management (CIKM 2026)

   PDF DOI

   Boukhers, Z., Kong, L., Zabulis, X., & Toubekis, G. (2026). Automated Construction of FAIR Digital Object Knowledge Graphs from Flat Cultural Heritage Records. Proceedings of the 35th ACM International Conference on Information and Knowledge Management (CIKM 2026). https://doi.org/10.1145/3799682.3840068

   Boukhers, Zeyd, , and Lingxiao Kong, , and Xenophon Zabulis, , and Georgios Toubekis. "Automated Construction of FAIR Digital Object Knowledge Graphs from Flat Cultural Heritage Records." Proceedings of the 35th ACM International Conference on Information and Knowledge Management (CIKM 2026), 2026.

   Boukhers, Zeyd, , and Lingxiao Kong, , and Xenophon Zabulis, , and Georgios Toubekis. "Automated Construction of FAIR Digital Object Knowledge Graphs from Flat Cultural Heritage Records." Proceedings of the 35th ACM International Conference on Information and Knowledge Management (CIKM 2026) (2026).

   @inproceedings{Boukhers2026automated,
     title = {Automated Construction of FAIR Digital Object Knowledge Graphs from Flat Cultural Heritage Records},
     author = {Zeyd Boukhers and
              Lingxiao Kong and
              Xenophon Zabulis and
              Georgios Toubekis},
     booktitle = {Proceedings of the 35th ACM International Conference on Information and Knowledge Management (CIKM 2026)},
     year = {2026},
     doi = {10.1145/3799682.3840068},
     url = {https://doi.org/10.1145/3799682.3840068},
     publisher = {Association for Computing Machinery},
     
   }

   Cultural heritage records frequently represent entities such as people, places, materials, and periods as plain-text metadata values, limiting machine-actionability and cross-record navigation. This paper presents a pipeline for transforming Europeana records into a FAIR Digital Object knowledge graph structured with CIDOC-CRM. A large language model determines whether metadata values should become resolvable entity references or remain literals and links resolvable entities to controlled vocabularies such as Getty AAT, Wikidata, VIAF, and PeriodO. Evaluation on 637 archaeological records shows substantial automatic linking of previously unenriched metadata and cross-lingual entity consolidation.

View all publications

Upcoming Events & Talks

No upcoming events scheduled at the moment.

Check back soon for future talks and presentations!

View all events