“Trying to understand the world really drives me”

Nicolas Flammarion © 2021 EPFL

Nicolas Flammarion © 2021 EPFL

Nicolas Flammarion heads EPFL’s Theory of Machine Learning Laboratory (TML), becoming hooked by the potential and real-world impact of machine learning during his master’s degree. He recently won an Outstanding Paper award amongst almost 9-thousand that were submitted to the most influential conference in ML - NeurIPS.

Tenure-track assistant professor Nicolas Flammarion has always been driven by the idea of trying to understand the world and why things work the way they do. He recalls that even as a child, he loved science, “I really liked to understand how something worked and I remember that I was always disassembling the machines around my home but I was never able to put them back together. My family was both upset and amused by this!”

Fast-forward two decades and Flammarion’s work now focuses on some of the challenging problems around machine learning that are yet to be solved. “Machine learning is an interdisciplinary field at the intersection of many other fields, including optimization, statistics and computer science. It’s really fast moving and there are still big and fascinating outstanding questions to answer. For example, we have some algorithms with which we can obtain really superhuman performance, say in image classification, and in practice these work perfectly, but theoretically we barely understand how they do what they do,” he says.

Recently, a paper about optimization, A Continuized View on Nesterov Acceleration for Stochastic Gradient Descent and Randomized Gossip, co-authored by Flammarion, was selected for an Outstanding Paper Award by one of the world’s most prestigious machine learning conferences, NeurIPS; one of only six publications chosen among almost 9-thousand submitted. The paper takes a new approach to an old problem, Nesterov’s algorithm, trying to better understand how it works and offering a new, accelerated algorithm with a theory that is both understandable and that works in practice.

Flammarion says the accolade is an amazing achievement and points to both the success of collaboration and the growing prestige of EPFL in machine learning, “It’s very rewarding and really validates the research we are doing. I think it’s also important to point out that two of the authors who were students in Paris when we wrote the paper are now post-docs at EPFL and it shows that in machine learning we are attracting some of the best researchers and students. That makes me very happy and proud.”

Another aspect of his job is teaching. Flammarion jointly runs the Machine Learning course, the biggest master’s course on campus, and really enjoys keeping the content up to date in such a fast-moving field, “This year we introduced a new lecture on ethics and fairness which was really important to me. Previously, for example, a lot of people thought that a decision made by an algorithm must be neutral and fair, but algorithms are only as good as their input. I think it’s really important to talk to the students about this so they have it in mind when they start to work,” he says.

Flammarion joined EPFL in 2019 and, although much of his tenure has been clouded by the COVID-19 pandemic, he hasn’t looked back once, “I did my postdoc at Berkeley in California, one of the top universities for machine learning and it was very lively, so when I arrived at EPFL I had really high expectations. I’m really delighted to be here every day in such a vibrant work environment in such a beautiful town. I’m so happy that I chose to come to EPFL.”

Author: Tanya Petersen

Source: Computer and Communication Sciences | IC

This content is distributed under a Creative Commons CC BY-SA 4.0 license. You may freely reproduce the text, videos and images it contains, provided that you indicate the author’s name and place no restrictions on the subsequent use of the content. If you would like to reproduce an illustration that does not contain the CC BY-SA notice, you must obtain approval from the author.