News Publications and Preprints Notes Services
Mehrnaz Mofakhami

Hello! I am a Research Felow at Cohere, working on Multilingual Reasoning. Previously, I completed my master's at Université de Montréal and Mila with Gauthier Gidel and Ioannis Mitliagkas, and during my studies, I spent some time at ServiceNow Research as a Visiting Researcher. Prior to coming to Montreal, I received my bachelor's degree at Sharif University of Technology in Computer Science.

I am interested in the fundamentals of machine learning, where I can explore how AI can be used effectively and efficiently to improve decision-making. My research has evolved from studying predictive models that actively shape the environments they're designed to analyze to designing techniques uncovering hidden vulnerabilities of large language models and making them more robust. I'm currently focused on multilingual large reasoning models, to ground reasoning in as many languages as possible, as part of the Tiny Aya effort at Cohere Labs.

If there's anything you'd like to discuss with me, feel free to email me at: mehrnaz (dot) mofakhami [at] mila.quebec or reach out on X!

profile photo

News

August 2024
I'm excited to announce that I'm co-organizing the Women in Machine Learning (WiML) workshop at NeurIPS 2024 as the Mentorship and Networking Program Chair.
July 2024
I presented my work on "Performance Control in Early Exiting to Deploy Large Models at the Same Cost of Smaller Ones" at the ES-FoMO workshop at ICML 2024 in Vienna.
May 2024
I am very grateful to have received the AI Scholarship for Excellence in Research from the University of Montreal's Graduate and Postdoctoral Studies.
March 2024
Happy to have received a full travel grant to attend the Cornell, Maryland, Max Planck Pre-doctoral Research School in Computer Science (CMMRS) in Summer 2024.
October 2023
I started an internship with ServiceNow Research at the Multimodal Foundation Models team.
Fall 2023
I got the merit-based Excellence Scholarship from DIRO (Department of Computer Science and Operational Research) at the University of Montreal for the third time in a row!
June 2023
I presented an introduction to the Performative Prediction framework and my research on this topic at a Google DeepMind Montreal Tea Talk.
September 2022
I joined the Space Committee at Mila to guide decisions surrounding the management of the space for students and profs.

Publications and Preprints

Building Multilingual Bridges: Data Mixing as the Pillar of Generalization for In-Language Reasoning
Mehrnaz Mofakhami, Ananya Sahu, Alejandro R. Salamanca, Daniel D'souza, Alexandre Berard, Thomas Euyang, Marzieh Fadaee, Julia Kreutzer
Preprint, 2026
Paper / Model / Data
Tiny Aya: Bridging Scale and Multilingual Depth
Alejandro R. Salamanca, Diana Abagyan, Daniel D'souza, …, Mehrnaz Mofakhami, et al.
Alejandro R. Salamanca, Diana Abagyan, Daniel D'souza, Ammar Khairi, David Mora, Saurabh Dash, Viraat Aryabumi, Sara Rajaee, Mehrnaz Mofakhami, Ananya Sahu, Thomas Euyang, Brittawnya Prince, Madeline Smith, Hangyu Lin, Acyr Locatelli, Sara Hooker, Tom Kocmi, Aidan Gomez, Ivan Zhang, Phil Blunsom, Nick Frosst, Joelle Pineau, Beyza Ermis, Ahmet Üstün, Julia Kreutzer, Marzieh Fadaee
COLM, 2026
Paper / Blog post
The Culture Funnel: You Can't Align What Isn't in the Data
Ananya Sahu, Mehrnaz Mofakhami, Daniel D'souza, Thomas Euyang, Julia Kreutzer, Marzieh Fadaee
Preprint, 2026
Paper
A Coin Flip for Safety: LLM Judges Fail to Reliably Measure Adversarial Robustness
Leo Schwinn, Moritz Ladenburger, Tim Beyer, Mehrnaz Mofakhami, Gauthier Gidel, Stephan Günnemann
ICML, 2026
Paper
A generative approach to LLM harmfulness detection with special red flag tokens
Sophie Xhonneux*, David Dobre*, Mehrnaz Mofakhami*, Leo Schwinn, Gauthier Gidel
BuildingTrust Workshop, ICLR 2025 | * Equal Contribution
Paper
Performance Control in Early Exiting to Deploy Large Models at the Same Cost of Smaller Ones
Mehrnaz Mofakhami, Reza Bayat, Ioannis Mitliagkas, João Monteiro*, Valentina Zantedeschi*
Efficient Systems for Foundation Models Workshop, ICML 2024 | * Equal Supervision
Paper
Tight Lower Bounds and Improved Convergence in Performative Prediction
Pedram Khorsandi, Rushil Gupta, Mehrnaz Mofakhami, Simon Lacoste-Julien, Gauthier Gidel
NeurIPS, 2025
Paper
Performative Prediction on Games and Mechanism Design
António Góis, Mehrnaz Mofakhami, Fernando P. Santos, Gauthier Gidel, Simon Lacoste-Julien
AISTATS, 2025
Paper / Code
Performative Prediction with Neural Networks
Mehrnaz Mofakhami, Ioannis Mitliagkas, Gauthier Gidel
AISTATS, 2023
Paper / Video
Reproduction: Adversarial Example Games
Adversarial Machine Learning course
Report / Code
Reproduction: Tracking the World State with Recurrent Entity Networks
Mehrnaz Mofakhami, AmirHossein Yavari
EEML Summer School 2021 - Best poster award
Report / Code

Notes

Tutorial: An introduction to Robust and Trustworthy ML
Supplementary material for the Artificial Intelligence Course at SUT - Spring 2021

I wrote this short tutorial while I was a TA in the AI course at Sharif University of Technology. It is an introduction to the main topics in robust and trustworthy ML, including evasion and poisoning attacks, and mechanisms to defend against them.

Adversarial learning course - IFT 6164: Adversarial Examples: part 2

This scribe note is based on the lectures of Professor Gauthier Gidel in Adversarial Machine Learning course , Winter 2022.

Services