İsmail İlkan Ceylan
Principal Investigator
Generative Modeling
Flow matching and related generative models for discrete, geometric, and topological data.
[TODO: review this text.]
Generative models are increasingly used to design molecules, materials, and sequences. We develop flow-based generative models that respect the structure of the data they model, from discrete sequences to geometric and topological objects.
Key Directions
- Discrete generative modeling: Flow matching over discrete and categorical data, including few-step generation with flow maps.
- Geometry and topology: Flow matching on spaces with topological structure, and topological guidance for generation.
- Dynamics: Learning non-gradient dynamics and couplings for guided flows.
People
Alicja Maksymiuk
PhD Student
Angelo Gnazzo
PhD Student
Kacper Wyrwal
PhD Student
Katarina Petrović
PhD Student
Oscar Davis
PhD Student
Ayhan Suleymanzade
External Collaborator
Selected Publications
All publications →Thinking with Looped Flows
In arXiv preprint
September 2026
Categorical Flow Maps
In ICML 2026
July 2026
MacroGuide: Topological Guidance for Macrocycle Generation
In ICML 2026
July 2026
Topological Flow Matching
In ICLR 2026
April 2026
Curly Flow Matching for Learning Non-gradient Field Dynamics
In NeurIPS 2025
December 2025
Global Transport Couplings for Classifier-Free Guided Flows
In arXiv preprint
October 2026