Learning (Approximately) Equivariant Networks
ACE lets a network explore during training, then adaptively enforces exact or approximate symmetry.
Read spotlightMLS Lab · University of Stuttgart
Spotlights
Our papers in a nutshell: the ideas, methods, and results behind our research.
Geometric and structure-aware learning
Generative and probabilistic models
Logical Guidance for the Exact Composition of Diffusion Models
LOGDIFF turns AND, OR, and NOT into probability-aware guidance, preserving multiple valid outcomes during diffusion.
Read spotlightMachine learning for science and engineering
Distillation of Foundation Models for Time-dependent PDEs
TREX extends a few simulations with a large teacher, creating richer training data for a compact, fast student.
Read spotlight