François Gonard joined SystemX in November 2014 to write a thesis on the topic “Cold-start recommendation: from Algorithm Portfolios to Job Applicant Matching”. He returns to his Ph.D. carried out within the project Model Reduction and Multi-Physical Optimization (ROM) and supervised by Inria.


What is the subject of your thesis?

Manufacturers are increasingly using optimization in their car part or reactor design processes, for example. It is a question of finding the best values of thickness or angle for a criterion (such as the mass in aeronautics) while satisfying the previously fixed constraints for the part (maximum deformation in a point for example). The basic idea of my thesis is that there are several methods (optimization algorithms) to carry out this task and that each one works more or less according to the problem (the combination part-criterion-constraints) that one must solve. It is therefore a question of finding the most suitable method from the definition of the problem.
For this, I use machine learning to learn the “problem – method to choose” relation on a database of known problems. When a new problem needs to be solved, the system I designed proposes the algorithm to use, or even a combination of algorithms. More precisely, it is composed of two components. The scheduler solves easy problems, that is to say those for which there is at least one algorithm that solves them in a very short time. We will execute a sequence of well-chosen algorithms (from our database) for a short time each. The second component, the selector, is dedicated to solving difficult problems: it predicts the problem solving time by each of the algorithms and selects the one with the best resolution time estimated.
Subsequently, I worked on the problem of automatic matching of job ad and seekers. It is quite similar: for each candidate, there are job ads that are more or less suitable. We will therefore make personalized recommendations, again using a system that learns the “candidate – tailored ad” relation on a database.

What do you remember from your doctorate?

The thesis is an adventure with ups and downs and a lot of personal involvement. I am happy to have discovered new areas, I learned a lot and I was able to advance some topics. In the end, it is the pride of being recognized for my work that takes precedence.

Can you tell us about your best memories at SystemX?

I have very good memories of my participation in the 2017 edition of the Paris-Saclay Basketball Trophy. We had set up a team for the occasion, the #TeamSystemX, to compete against teams from other businesses and schools in the Plateau de Saclay.

What are your plans for the future?

I have just been hired as Data Scientist at Sevenhugs, a start-up that develops connected objects. I will continue to exploit data, but for applications different from those that I worked on in my thesis.


Find out more about François Gonard
Thesis Subject: Cold-start recommendation: from Algorithm Portfolios to Job Applicant Matching
R&D Project: Réduction de modèles et Optimisation Multi-physiques (ROM)
Last degree before PhD: IngénieurSupaero


Subscribe to IRT SystemX's

and receive every month the latest news from the institute: