The Interplay of Data Structure and Imbalance in the Learning Dynamics of Diffusion Models
Real-world datasets differ across classes in both structure and frequency, but most theory for diffusion models assumes homogeneous...
A theory lab forunderstandinglearning systems.
We use statistical physics, probability, and computer science to understand why learning systems generalise, fail, and develop bias, and how to build safer and more intelligent AI.
Explore our researchOur latest works
Position: the Stochastic Parrot in the Coal Mine. Model Collapse is a Threat to Low-Resource Communities
Model collapse can degrade generative models when they are trained on outputs from earlier models. This position paper...
News
Flavio and Chenxiao will present our preprint on how data structure and imbalance shape learning in diffusion models at the Learning Theory Summer School workshop in Copenhagen.
Stefano will give a talk at KITP's New Trends in Non-equilibrium Dynamics program, presenting our ICML 2026 paper on the loss landscape of overparameterised two-layer ReLU neural networks.
Today starts the Mathematical Foundations of AI workshop, organised by Flavio and Stefano. The programme brings together diffusion models, transformers, and associative memories, and includes a contributed talk by Chenxiao on our work on heterogeneous data in diffusion models.
Great news! Stefano has been awarded 39,000 SEK from Helge Ax:son Johnsons stiftelse, supporting research travel and international scientific exchange.
People across disciplines
A small group asking questions.
We work across machine learning, statistical physics, cognitive science, and high-dimensional probability at Chalmers and the University of Gothenburg, with active links to Wits.
Chenxiao Ma
PhD Student
Flavio Nicoletti
Postdoc Researcher
Loek van Rossem
PhD student
Stefano Sarao Mannelli
Assistant Professor
Meet the group