Howdy!! I'm a PhD student at Brown University, where I'm fortunate to be advised by Chen Sun. I'm supported by Brown's Department of Mathematics and Department of Computer Science. I study machine learning, computer vision, and artifical intelligence. My current projects focus on generative modeling, applied to various domains. In particular, I like building generative models with mathematical or physical inductive biases that make their outputs more realistic. If you're at Brown and you're interested in working on a research project with me, that's awesome!! Please email me directly, and attach your CV and transcripts.

In the past I also did work in cryptography and pure mathematics, including number theory, algebraic geometry, and geometric measure theory. Fun fact: I actually started grad school as a PhD student in Brown's math department, conducting research in analytic number theory and cryptography with Jeff Hoffstein. I've since switched to AI, but I still like to make my background in pure math useful in my AI research. After getting my masters degree in mathematics in spring 2022, I took a professional leave of absence for a year to gain exposure to ML in industry. I did three internships: at American Express AI Labs , Akkio (a no-code AI startup), and Captions (an AI video editing startup).

I completed my undergraduate degree at Wesleyan University. During my time in college I spent one semester at the Math in Moscow program and another at the Budapest Semesters in Mathematics program. My undergraduate math research advisor was Ken Ono, I spent two summers doing research with him at Emory University's Research Experience for Undergraduates.

I'm particularly inspired by the life of Walter Pitts, who proposed the first mathematical model of the neural network.

Resume  /  Google Scholar  /  Semantic Scholar  /  Github  /  arXiv  /  LinkedIn  /  Twitter

nate_gillman [at] brown.edu

---

Publications (AI/ML)

2024 2022

Patents (AI/ML)

2022

Publications (Mathematics)

2021 2020 2019 2018

---

Updates

• [Oct-2024] Our arXiv preprint proposes a new neural network layer, the Fourier head, which learns a continuous probability density function using Fourier series, and returns a discrete approximation of it. When to use it? Large language models are often adapted to model non-linguistic tokens. If these tokens have an underlying continuous structure, then replacing the linear classification head with the Fourier head can boost downstream performance. You can find code and visuals on our project page.
• [Feb-2024] Our arXiv preprint aims to stabilize self-consuming generative model training. We support our proposed method with rigorous proofs, as well as experiments on the challenging human motion synthesis task. You can find human motion visuals and code on our project page.