Sites Inria

Version française

Inria Awards 2012

Isabelle Bellin -

Young researcher Inria award: Francis Bach

Francis Bach - © Inria / Photo J.M. Ramès

A trained mathematician, Francis Bach has become an internationally acknowledged specialist in the field of statistical machine learning. A field he has been boldly exploring for some ten years now. His research at the interface between applied mathematics, statistics, and computer science finds applications in an impressive variety of domains, such as artificial vision, processing of audio signals, bioinformatics, and cerebral imaging. 

When Bach’s wife offered him a Backgammon set, she certainly had no idea she was about to help him find his calling. Nevertheless, as a result he understood, or rather had to admit to himself, that deep down he was a researcher. At the time, this former graduate of X-Mines was destined for an administrative or industrial career, in line with his prestigious academic background. Francis Bach maintains his interest for practical problems. In his research work, he particularly appreciates being able to move between different activities, from theoretical mathematic developments to discussions with colleagues about more hands-on challenges, then on to the design and programming of algorithms.

Backgammon opened his eyes! “The penny would have dropped sooner or later,” he admits, “it was unavoidable.” During a trainee period in a company, being of a curious nature, he endeavored to understand the famous strategy game, especially how an algorithm could beat the world champion. He explored the scientific literature on machine learning theory and spent his days implementing the algorithms. “I knew little about computer science, I could only program using C language, but I was immediately thrilled.” He completes a thesis at Berkeley about machine learning (also known as automatic learning) under the direction of Michael Jordan, one of the stars in the field, professor of computer science and statistics. “A wonderful experience,” he says, “where everything is open to question (over there X-Mines graduate engineer means nothing). Extreme dynamism, a collective desire to surpass, a degree of confidence that gives wings.”

At Berkeley, he discovers a topic that shall remain his field of research: machine learning for the purpose of developing automated processing methods, the only methods capable of handling the complex and increasing amount of digital data surrounding us: images, texts, sounds for multimedia applications, imaging or genomic data for medical applications, etc. These methods are crucial for the optimisation, monitoring and modeling of complex systems based on examples. Francis Bach also discovers a working method that shall be his guiding principle: the search for methodological tools that can be shared by the different applications. “We often miss the years spent in Berkeley. I have been lucky in finding a similar work environment in the ENS Paris laboratory and in Jean Ponce’s Inria team working on artificial vision. Had it been different, I would certainly have returned to the United States,” he admits. He particularly appreciates the research conditions, the students he works with, and the positive dynamism.

In his research on automatic processing of massive and complex data, he easily moves between applications, theory and algorithms. His objective is to find solutions with reasonable computation times. He always starts with data from real problems. By searching for similarities, for example temporal ones, between a DNA sequence and a sound sequence, he develops subtle algorithms based on these properties, which until now were rarely used, with, as a corollary, algorithms adapted to several applications. This search for new solutions for the initial processing of data touches on domains such as artificial vision (recognition of scenes and objects, image noise reduction) and audio signal processing (separation of sound sources, music recognition). “I believe this upstream work is important, as in the end we are manipulating the same type of objects, for example three-dimensional sequences or objects,” he concludes.

“This search for shared methodological tools is also a core theme of my European Research Council (ERC) scholarship,” he says. This scholarship was awarded in 2009 for the purpose of developing generic methods while taking into account the physical constraints of a prediction problem. Example: processing massive but known data concerning few cases such as the genomes of a few people only. Classical statistics is too generic and inoperative. “We are looking for methods for finding a limited group of data on which we can make statistics,” he summarises. In particular, this concerns parsimonious methods, which were also the subject of his post-doctoral thesis, which he defended the same year.

He no longer plays much Backgammon. His playing field has grown enormously, both in terms of theory and application. He may have taken his time to find his calling, completing his thesis at the age of 30, but since then he has been making progress in leaps and bounds.


Jean Ponce, director of the computer science laboratory of the Ecole normale supérieure (Paris), manager of the Willow project team, Inria Paris-Rocquencourt

Francis Bach has created a real dynamic in France around machine learning. I heard about him from colleagues when I was in the United States

Jean Ponce

myself. So when I was creating my team to launch the Willow project, I went to see him. I was impressed by his intelligence, enthusiasm and intellectual curiosity, and I was delighted he wanted to join us, contributing his learning expertise to the team. Nothing has tarnished this first impression. Francis is certainly one of the best researchers of his generation in the domain of machine learning and its applications, in particular image processing and artificial vision. It is a pleasure working with him and he is always generating an abundance of original ideas. His presence has enabled us to recruit very good young researchers, either students or already experienced researchers. I was also impressed by the quality of the group he brought together for his own team. The time had come for him to achieve this well-deserved independence, both within Inria and his ERC scholarship.”

Michael I. Jordan, professor at the computer science department of the University of California (Berkeley), Francis Bach’s thesis advisor

Michael I. Jordan - DR

“It is really quite simple: in my 20-year career at Berkeley and at the Massachusetts Institute of Technology, I have had no better doctoral student. Francis is exceptional, multidisciplinary, and capable of attacking the most complex problems with a very personal approach, a unique manner of using algorithmic statistics. He develops original solutions, which at first meet with resistance, then with unanimous approval. In fact, his algorithms are precise and robust, be they for speech recognition like for his thesis or for the vision domain. Francis benefits from his excellent French training in applied mathematics. He is highly skilled in the domains of optimisation, linear algebra and signal processing. Very quickly during his doctoral thesis, he brought together a group of students to work on his ideas. In my opinion, his current team at Inria is one of the best in the world. French research will surely benefit from this dynamism. In 10 years, Francis Bach has become one of the world leaders of his generation in machine learning.”

Keywords: Inria Award Young researcher award Francis Bach