My machine learning group will launch at UW-Madison this Fall 2026!
My group's aim is to design algorithms that generate and learn from data. An emphasis will be placed on developing a mathematical understanding of when these algorithms scale, and on collaborating with domain scientists to deploy them on real data. Some directions we plan to explore are:
- diffusion models for text data: discrete token spaces, variable-length sequences, and product spaces combining text with other modalities
- sampling algorithms for molecular data: scalable methods for multimodal landscapes and representations that aid sample generation
- causal discovery algorithms for brain imaging data: methods that scale across subjects and help characterize stimulus-response pathways in dynamical systems
- theory for when machine learning algorithms scale: sample or computational complexity as a function of design choices
Joining my group
Prospective Ph.D. students
I anticipate hiring 1-2 PhD students to join me at UW-Madison starting in Fall 2027. Ideal candidates will have a strong foundation in applied or computational mathematics, statistics, or engineering and be proficient in at least one programming language. Experience with chemistry or brain imaging is useful, but is not required. To be considered, apply through UW-Madison's Electrical and Computer Engineering Ph.D. program and mention my name in your application.
Prospective research interns
I am recruiting research interns in the areas listed below. These projects are generally aimed at master's and Ph.D. students, but advanced undergraduates are welcome to apply. Students at other U.S. institutions or at universities abroad may be considered for remote projects.
- Codebase for Causal Discovery in Brain-Imaging Datasets
- Training Energy-Based Models on Discrete Data, e.g. quantized images and text
- Optimal Transport Guarantees for Discrete Generative Models
- Mathematical Analysis of Tempered Sampling Algorithms
- Statistical Foundations of Modern Representation Learning
- Statistical Estimation Theory for Modern Generative Models
Applicants may also propose a related direction of their own.
