MLS Lab · University of Stuttgart

Generative and probabilistic models

Developing methods for structured generation, probabilistic inference, and efficient sampling.

Overview

We develop generative models and probabilistic methods that incorporate structure and constraints. Our work introduces new model architectures, training objectives, and sampling algorithms for efficient and controllable generation.

Conceptual illustration of a cloud of blue points forming an ordered crystalline structure.

Research topics

  • Symmetry-aware generative models
  • Diffusion models and sampling methods
  • Probabilistic approaches to conditional generation

We design methods for one-shot generation under symmetries, gradient estimation for equivariant diffusion, and tractable conditional generation.

Selected publications

Information reviewed 26 September 2026 · University research topic