How to Join

For Prospective Students

Thank you for your interest in our lab. To learn more about our research, please explore the information provided on this website, in particular the Research Interests and Publications pages.

We hire graduate students from those who have completed an internship in our lab. If you would like to join as a graduate student, please apply for an internship first. The internship lets both sides find out whether our research style is a good fit before committing to a degree.

Undergraduate research opportunities are also available. If you are interested, you are welcome to contact us by email.

What to Expect

We solve problems caused by finite-precision arithmetic in machine learning, and our main tool is mathematics. Most of your time will be spent reading papers, formulating problems, proving theorems, and writing, rather than running large-scale experiments; our proofs typically use real analysis, probability theory, and combinatorial constructions. A typical project starts from a concrete open question, for example: what is the minimum width for a network to be a universal approximator, or what does automatic differentiation actually compute for a network running on floating-point arithmetic.

Research from the start. Students work on research problems from early on rather than after years of coursework. Both of our current students are co-authors of an ICML paper: see, for example, our ICML 2025 paper on the minimum width for universal approximation.
Close guidance. The lab is small by design, and you will discuss your problem with the advisor regularly and in depth, from choosing the question to polishing the final write-up.
Publishing. Our results appear at machine learning venues such as ICML, ICLR, NeurIPS, and COLT. You will learn to write a rigorous paper and to present your work.
Internship. Interns are treated as junior researchers: you will be given a concrete problem, learn the necessary background, and try to make progress on it together with us. Successful internships often turn into a paper and into an offer to join the lab.

Recommended Background

A strong background in mathematics and algorithms is the best preparation for our lab. In particular:

Required: undergraduate-level linear algebra, probability theory, and computer programming.
Helpful but not strictly required: courses in optimization, mathematical statistics, or theory of computation, and familiarity with machine learning theory.

Prior research experience is not required. What matters most is mathematical maturity: the ability to work through a proof carefully, to notice when an argument is incomplete, and to keep going when a problem does not yield quickly.

Contact

If you are interested in an internship or in joining the lab, please send an email to the address on the Contact page, with your CV, your transcript, and a brief description of your research interests. To show that you have read this page, please include the word floating-point in the subject line of your email.