Dr. Sebastian Sanokowski

Google Scholar

Research Interests

I conduct research at the intersection of generative modeling, combinatorial optimization, and sampling. My focus is on developing diffusion-based methods that enable learning and sampling in complex discrete spaces without relying on labeled data. This includes applications in optimization, statistical physics, and shape generation under geometric constraints. I’m particularly interested in building scalable and theoretically grounded approaches that remain robust across diverse tasks and domains. My work has been published at top machine learning conferences such as ICML, NeurIPS and ICLR. 
Overall, my goal is to design models that combine strong theoretical foundations with practical utility.

Selected Publications:
[1] Sanokowski, Sebastian, Sepp Hochreiter, and Sebastian Lehner. "A Diffusion Model Framework for Unsupervised Neural Combinatorial Optimization." ICML 2024.
[2] Sanokowski, Sebastian, et al. "Scalable Discrete Diffusion Samplers: Combinatorial Optimization and Statistical Physics." ICLR 2025.
[3] Sanokowski, Sebastian, et al. "Variational annealing on graphs for combinatorial optimization." NeurIPS 2023.
[4] Sanokowski, S., Gruber, L., Bartmann, C., Hochreiter, S., & Lehner, S. Rethinking Losses for Diffusion Bridge Samplers. NeurIPS 2025.
[5] Berzins, A., Radler, A., Volkmann, E., Sanokowski, S., Hochreiter, S., & Brandstetter, J. Geometry-Informed Neural Networks. ICML 2025.

Education

I earned my PhD in Artificial Intelligence from Johannes Kepler University Linz, where I completed my thesis on "Data-Free Combinatorial Optimization Using Generative Neural Networks" under the supervision of Prof. Sepp Hochreiter. 
I hold a Master of Science in Physics and Bachelor of Science in Physics from the Friedrich-Alexander-Universität Erlangen-Nürnberg.