UCLA · Since 2024

BruinML

A machine learning and mathematics research group at UCLA.

Random search + gradient ascent for nonconvex optimizationfive Gaussian peaks

Both methods run gradient ascent from the same random starts. Multistart pays to evaluate f at every endpoint. LipScreen first checks a Lipschitz bound and skips any endpoint that can't beat its best so far.

Multistart0 f-calls
LipScreen0 f-calls
gradient path evaluated best ×skipped ruled out
run 0/40
Open the full 3D demo →

Who we are

BruinML is an undergraduate research group at UCLA, founded by William Chang in January 2024. We started with a handful of students working on multiplayer bandits and reinforcement learning, and have since grown to more than 100 students. Our past presidents are Kenny Guo (2024–2025) and Taro Iyadomi (2025–2026).

Our philosophy

We believe research ability is cultivated, not screened for. BruinML is one of the few research programs in the country with no admissions process, and we think our results show it works: everyone who comes to us gets a real chance at a real project. What students make of that chance is up to them, and the ones who show up, ask questions and keep moving are the ones who see their names on a paper.

Nobody is expected to arrive with the background already in hand. We bring students up to speed with consistent mentorship and a shared set of resources, from reading lists and paper templates to a group workstation for experiments, so that each person has what they need to succeed on their project.

What we work on

Most of our work is in machine learning theory: multiplayer and multi-agent bandits, reinforcement learning in continuous time and with decentralized agents, optimization on manifolds and nonconvex landscapes, equilibria in games, Shapley-value explainability, and accountability in multi-agent LLM systems. We also work in applied mathematics, from CT image reconstruction for limited-angle breast tomosynthesis to optimal gaze control and nonlocal discrete operators, and we keep branching into new problems in both fields.

How a project runs

Each project is a team of three to five students. In weekly Zoom meetings led by William, the team reads a paper that interests us, learns the background it needs, and builds an open problem of its own within that area. Once the team has a problem and the tools to attack it, the meetings give way to independent work, and the project ends as a paper submitted to a conference or journal.

We recruit most of our members through the UCLA Directed Reading Program, which we have taken part in nearly every quarter:

  • Fall 2025
  • Winter 2026
  • Spring 2026
  • Fall 2026

Who leads it

William Chang is a PhD student in Applied Mathematics at UCLA. His research looks for theoretical understanding of physical phenomena, a category in which he counts machine learning itself.

In their words

I learned that research is not simply about completing assigned tasks, but about taking the initiative to pursue questions that remain unanswered by humanity.
Min Park, Applied Mathematics, UCLA ’28

Our students and alumni wrote about what doing research here is actually like, including the hard parts.

What we’ve done

Our students publish at machine learning and mathematics venues every year. If that sounds like something you’d like to do, come join us.

100+Students since 2024
–Papers on this site
–Co-authors
–Research topics

See the BruinML resume (every paper on one sheet, also as a PDF)

Research

Loading papers…

TopicsPick one to see the papers

    Join us

    We're always looking for curious students. If you like math, coding or both and want to try real research, we'd love to hear from you.

    1. Send us an emailTell us a bit about yourself: your year, your major, the courses you've taken and what you're curious about.
    2. Find a projectWe'll match you with a mentor and a problem that fits your background, often alongside other students.
    3. Write the paperWork through the theory and experiments, then write up and submit to a conference or journal together.
    All papers