My primary area of research is generative modeling, with a focus on understanding the geometry of high-dimensional data as captured by diffusion and flow matching models and using these insights to improve them. For example, I have worked on estimating the local intrinsic dimension of data manifolds using pre-trained diffusion models, leveraging these estimates to detect out-of-distribution points or to identify when a diffusion model is memorizing.

Beyond generative modeling, I have also worked on causal inference, in-context learning, and training tabular foundation models.