Ioannis (Yannis) Mitliagkas

Ioannis Mitliagkas

Γιάννης Μητλιάγκας

I study optimization, dynamics and learning in modern machine learning: why training works, when models generalize, and how to make both reliable.

Ioannis Mitliagkas (Γιάννης Μητλιάγκας)
CV Email Scholar

Prospective students: Fall 2027

I am recruiting PhD and MSc students for Fall 2027. Please go over my recent publications and the research themes below. If you think we have a strong overlap in interests, submit a Mila supervision request between October 15 and December 1, 2026, and list me as one of your faculty of choice. You also need to apply separately to the Université de Montréal MSc or PhD program.

I cannot respond to all emails; the supervision request is how you will be considered. I will consider all good candidates, but pay extra attention to those from unusual backgrounds, underrepresented groups, and candidates coming from regions under threat of war, occupation, political instability, etc. (Ukraine, Palestine, Africa, …).

Optimization for modern deep learning

Why optimizers like Adam work, and how to make training faster and more robust: batch sizes, sign and spectral methods, distributed and asynchronous training.

Generalization, out-of-distribution robustness and compositionality

Models that keep working when the data changes: domain shift, OOD detection, identifiable and compositional representations, and pretraining objectives.

Privacy and machine unlearning

Removing the influence of data from trained models, with guarantees, and measuring whether unlearning methods actually work.

Dynamics of games and reinforcement learning

Min-max optimization, variational inequalities and the learning dynamics of multi-agent and reinforcement learning.

News

All news

Group

Members and alumni

My most important responsibility is supervising a group of very talented junior researchers.

Where alumni went next

About

Researcher in machine learning. Academic, immigrant, amateur musician, runner.

Every fall, I teach Fundamentals of Machine Learning (IFT 3395/6390, in French) to a large class of undergraduate and graduate students. In winter 2027 I am teaching my advanced research class on deep learning theory (IFT 6169) again.

I co-founded and hosted the first two seasons of MTL MLOpt, a bi-weekly meeting of optimization experts from Mila, UdeM, McGill (CS and math), Google DeepMind, SAIL, FAIR and MSR. We share our guest speaker videos.

In the early days of interest in the area, I co-organized the Smooth Games Optimization and ML workshop series at NeurIPS. The opening remarks from NeurIPS 2019 summarize our motivation. In spring 2022 I was invited to the semester on Learning and Games at the Simons Institute, Berkeley. For several summers I taught optimization for ML at the Neuromatch Academy deep learning course.

Before joining the Université de Montréal, I was a postdoc with the Departments of Computer Science and Statistics at Stanford University, and a PhD student at The University of Texas at Austin.