Jan Ramon: machine learning, at the heart of research

You joined Inria in 2015 and became a member of the Magnet project-team. What is the aim of your work?

Magnet is the acronym for machine learning in large information networks . Our project-team is a joint unit of the CNRS, Lille 3 and Lille 1 Universities* consisting of 14 researchers. We are working on defining methods and models for automatic machine learning within data networks. Data collection can concern very different fields, from traditional networks such as highway infrastructures or economic networks, to social or biological networks, etc.

What areas of activity are impacted by your research?

Our work may prove useful for standby systems or data mining/extraction as well as for recommendations or predictions as regards behavioral patterns. Medical biology is an important field of application. Based on a patient’s historical data, we can establish predictions about their medical future and determine, for example, what types of complications could occur during hospitalization or what kind of drugs are liable to provide a cure.

What kind of stakeholders are you targeting?

Our research is of interest to many different stakeholders, both in the academic field and in industry. In general, these stakeholders are in possession of enormous masses of data that they are looking to exploit in order to improve their operating processes. The applications are very extensive: in the same way that we can foresee the future for human patients, data processing makes it perfectly possible to predict potential failures in the facilities or machinery in a production plant. From a mathematical point of view, the warning process that precedes the risk of a problem is the same for human patients and for machines.

You have just obtained an ERC (European Research Council) Proof of Concept Grant for the SOM project. What does that mean for you?

SOM is based on the MiGraNT (Mining Graphs and Networks) research project which concluded in 2015 and for which I received of an ERC Starting Grant at KU Leuven in Belgium. In the framework of MiGraNT we developed an approach to data mining based on information represented in the form of graphs. Our work led to the development of a theoretical understanding of learning in respect of building algorithms usable in applications for everyday life.

What is the difference between SOM and MiGraNT?

The ERC-PoC SOM project will allow us to put the theory developed within MiGraNTinto practice.We are going to be reaching out to communities and businesses to convince them that the concrete application of these projects can bring them economic benefits.

How would you define the concrete applications of the SOM project?

The purpose of the project is to investigate learning methods that are capable of defending the confidentiality of data structured in networks. We are especially interested in areas involving mobile devices. Examples are crowd sourcing or maybe ride sharing, where we will try to predict the possible route of a driver based on data that they have input into the application previously. The idea is to improve the predictive character of certain modes of data processing while guaranteeing compliance with confidentiality rules.

What study projects are you involved in at Inria?

At this time, I am working on the MUST project, which aims to reduce CO2 emissions on our roads. We are working in partnership with the manufacturers of devices to record the GPS locations of around 2,000 vehicles. The idea is to analyze this data and understand how we can reduce gas emissions by, for example, following different routes.

* Joint Research Unit 9189 - CNRS-Centrale Lille-Université Lille1, CRIStAL.