İsmail İlkan Ceylan will join TU Wien as a Faculty Member and serve as an Adjunct Principal Investigator at AITHYRA. He also holds an academic affiliation with the University of Oxford, where he previously was an Assistant Professor (Departmental Lecturer) in Computer Science. He earned his Ph.D. in Computer Science from TU Dresden in 2017, with a dissertation that received several distinctions, including the E.W. Beth Dissertation Prize.

Dr. Ceylan’s research focuses on AI and ML, particularly graph machine learning, which explores learning from complex relational structures such as graphs and knowledge bases. His long-term goal is to develop reliable and robust learning systems capable of reasoning over relational patterns across diverse domains. This research draws on techniques from machine learning (e.g., foundation models, graph neural networks, geometric deep learning) and theoretical computer science (e.g., logic, probability, graph theory, descriptive complexity).

He has published extensively in top-tier venues including NeurIPS, ICML, ICLR, AAAI, IJCAI, KR, and Artificial Intelligence, earning best paper awards (KR, ICDT) and reviewer honors (NeurIPS, ICLR, IJCAI). He is active in the research community as an area chair and senior program committee member for leading conferences. At Oxford, he co-organized the Learning on Graphs and Geometry (LoGG) seminar series and supervised numerous doctoral and master’s students. He received a Teaching Commendation for his course on Graph Representation Learning.

At AITHYRA, Ismail Ceylan will focus on developing trustworthy, interpretable, and scalable geometric deep learning methods, emphasizing applications in biomedicine and scientific discovery.

Orchid number: https://orcid.org/0000-0003-4118-4689
LinkedIn: https://www.linkedin.com/in/ismaililkan
Website: https://www.cs.ox.ac.uk/people/ismaililkan.ceylan

What is the solubility of a molecule? How do certain genes interact with diseases? What movies might users prefer based on their profiles? How do proteins fold into their native 3D structures? How can we accurately estimate arrival times on road networks? In what ways can AI improve weather forecasting? Can we efficiently simulate complex phenomena like fluid dynamics?

These diverse and challenging questions share a common thread: they all require machine learning on structured, relational data—including graphs (e.g., conventional or geometric), knowledge bases (e.g., knowledge graphs, databases), or other relational representations. Such data is deeply embedded across domains—including the life sciences—and forms the backbone of many high-impact real-world systems.

The research of Ismail Ceylan focuses on advancing machine learning methods for relational data. Traditionally, this has involved developing and analyzing models such as Graph Neural Networks or Graph Transformers. More recently, his work has shifted towards foundation models for relational data—large-scale, pre-trained models that aim to replicate the success of Large Language Models (LLMs) in the graph domain. Unlike task-specific methods, these models are designed to generalize across tasks and domains, making them more suitable for real-world scientific and industrial applications.

A central goal of my research is to theoretically characterize the capabilities and limitations of existing methods—particularly in terms of expressiveness, generalization, and transferability—and use these insights to guide the design of novel architectures from first principles. This theory-driven approach helps us better understand the boundaries of current techniques and where innovation is most needed.

The ultimate goal of this agenda is to apply these next-generation models to high-impact scientific challenges, with a strong emphasis on accelerating scientific discovery by enhancing the interpretability, scalability, and reliability of graph-based machine learning systems—especially in domains such as biology, chemistry, and physics.

Publication Highlights

Xingyue Huang, Pablo Barceló, Michael M. Bronstein, İsmail İlkan Ceylan, Michael Galkin, Juan L. Reutter, and Miguel R. Orth. How expressive are knowledge graph foundation models? ICML 2025.

Linus Bao, Emily Jin, Michael Bronstein, İsmail İlkan Ceylan, Matthias Lanzinger. Homomorphism counts as structural encodings for graph learning. ICLR 2025.

Sam Adam-Day, Michael Benedikt, İsmail İlkan Ceylan, and Ben Finkelshtein. Almost Surely Asymptotically Constant Graph Neural Networks. NeurIPS 2024.

Xingyue Huang, Miguel R. Orth, İsmail İlkan Ceylan, Pablo Barceló. A Theory of Link Prediction via Relational Weisfeiler-Leman on Knowledge Graphs. NeurIPS 2023.

Ralph Abboud, İsmail İlkan Ceylan, Martin Grohe, and Thomas Lukasiewicz. The Surprising Power of Graph Neural Networks with Random Node Initialization. IJCAI 2020.