Geoffrey Négiar

Geoffrey Négiar

PhD Student

UC Berkeley, BAIR


I’m a 5th-year PhD student in UC Berkeley’s BAIR, co-advised by Prof. Michael Mahoney and Prof. El Ghaoui. My research focuses on leveraging mathematical optimization to incorporate and analyze structure in learning systems.

I’m currently a Student Researcher at Google Research, hosted by Fabian Predregosa, remotely in Montréal. I’m a member of the French Armament Corps (DGA, i.e. French DARPA Fellowship). I enjoy applying my research to industry problems: I’ve interned at Bloomberg LP, Shift Technology, and was a part time researcher at SumUp Analytics.


  • Machine learning
  • Optimization
  • Natural language processing
  • Cooking
  • Martial arts


  • PhD in Electrical Engineering and Computer Science

    UC Berkeley, BAIR

  • MSc in Machine learning (MVA), 2017

    ENS Paris-Saclay

  • Diplôme d'Ingénieur, 2017

    Ecole polytechnique


06/2021: I’m interning at Google Brain Montréal for the summer, hosted by Fabian Pedregosa! I’ll be working on optimization methods tied to program synthesis. Keep posted!

10/2020: We’ve open sourced CHOP, our optimization library built on Pytorch. CHOP contains methods for constrained and composite optimization, with a focus on 1) generating adversarial examples, 2) training sparse and structured neural networks.

06/2020: Our paper Stochastic Frank-Wolfe for Constrained Finite-Sum Optimization (G. Negiar, G. Dresdner, A. Y. Tsai, L. El Ghaoui, F. Locatello, R. Freund, F. Pedregosa) was accepted to the ICML 2020 main conference!

03/2020: I’m a finalist for the Two Sigma PhD Fellowship!

01/2020: Our paper Linearly Convergent Frank-Wolfe with Backtracking Line-Search (F. Pedregosa, G. Negiar, A. Askari and M. Jaggi) was accepted to AISTATS 2020!

What Is Machine Learning

In this post, I look at the different parts of a typical machine learning problem.

Plat en Sauce Abstraction

Hi all! Here is my first post on food! I hope you enjoy it and am looking forward to your comments! Today, I’ll focus on the French Plat en sauce concept.

Stochastic Frank-Wolfe

We present a novel Stochastic Frank-Wolfe method for finite-sum optimization.

Linearly Convergent Frank-Wolfe without Prior Knowledge

Structured constraints in Machine Learning have recently brought the Frank-Wolfe (FW) family of algorithms back in the spotlight. …


Making Frank-Wolfe algorithms practical and scalable.


  • ICLR 2021
  • NeurIPS 2021
  • ICML 2020


  • 2121 Berkeley Way, Berkeley, CA 94704
  • BAIR, 7th floor.
  • DM Me