Behind the scenes of research: imagining the future of network diagnostics

Joachim Flocon-Cholet, research engineer in data science, is co-presenter of the “Fault Diagnosis Discovery” demonstration at Orange’s 2021 Research Exhibition. He talks about the research process in business, its light-bulb moments and trial and error.

“To overcome frustration and use impatience as a driving force, you definitely need to be driven and persistent, and more than anything you need a healthy dose of creativity.”

The Research Exhibition marks a milestone in a project: it means that the project is now at a stage where it can be unveiled to the general public. Joachim Flocon-Cholet summarises things for the “Fault Diagnosis Discovery” demonstration as follows: when a customer calls an Orange call centre about a fault, the test launched on their line collects information on the network, the box, etc. A set of rules (“if”, “then”…) is associated with a list of known failures in order to look for matches. What if the problem can’t be identified? We call this an “Unknown Root Cause case”, which happens in 25% of tests. Wouldn’t it be enough to find new rules in this case? It’s not that simple. Hundreds of thousands of tests are generated by the tool every day. These in turn produce millions of pieces of data. Despite all their know-how, Orange business experts lack something crucial for solving such complex problems: time, a lot of time! This is where the research presented in the Fault Diagnosis Discovery demo comes in.

The birth of a vocation

For Joachim, who was already working in machine learning, things started to take shape when all the buzz around data started: i.e., Big Data. The Research Master’s student soon realised this field would become indispensable. He also saw career opportunities: the potential for a very wide field of applications, which touches on sectors as diverse as sound, image, agriculture, finance, and any other field that we can see is currently being revolutionised by data.

Acquired and unknown

At the root of the Fault Diagnosis Discovery project, there is a very clear research problem and a not so common object of study. “By starting to combine technological bricks and to test different algorithms, we realised that there was a problem with method. This is when we set up a method inspired by what is called ‘zero-shot learning’: our approach is to inject data taken from business knowledge to guide the system in the exploration and the discovery of unknown faults.” It was this exploratory aspect that drew Joachim to this research role four years ago. Many data projects start more from a controlled use case: we know what we want so we try to optimise the algorithm in this direction. With the fault diagnosis problem, however,  we don’t know exactly how to get to the end.

Experiments and more experiments

Forget about that “Eureka!” moment, because research is a test of endurance. “It takes several weeks to go from having an idea that works in theory and developing it, launching the calculations with synthetic data, then with real data, to finally presenting our work to be validated by our peers. There are the experimental phases, where I spend a lot of time coding, and the results consolidation phases, when I talk to my colleagues a lot. When you approach a publication, it’s the writing that takes precedence, and when I study a book to soak up new knowledge, I find myself lost in books and socially reclusive.”

Indispensable interactions

Isolation never lasts long, because the best way to stimulate thinking is to share your ideas. “What I miss most during lockdown and remote working are informal chats: passing colleagues in the corridor, discussing a concept, and going away for a couple of hours to scribble on boards… Interaction is key in the world of research. Besides, that’s why there are conferences.” In business, these interactions are also closely related to the reality on the ground. Joachim consolidates his progress with his research colleagues in the field of machine learning, but also with business experts, specialists in fibre optics for example. He continues to strengthen the link between research and application.

Sharp minds researching complex problems

“This is the type of work that takes so long you can sometimes lose sight of the common thread of the project. This is especially true when you aren’t making any progress, when you have to change everything. Fortunately, this is when you can rely on your team. To overcome frustration and use impatience as a driving force, you definitely need to be driven and persistent, and more than anything you need a healthy dose of creativity. We are often faced with problems that seem unsolvable. You have to want to solve them, you have to build countless scenarios in your head, try new things or different paths.”

This is how, from a small idea, from an intuition, we have come up with this new proposal – capable of durably transforming the fault diagnosis systems on the network, for obvious gains both on the operational side and in terms of customer experience.

Read also on Hello Future

NORIA: Network anomaly detection using knowledge graphs


FoodProX uses machine learning to detect ultra-processed food

GettyImages - machine learning - research recherche

Machine-Learning-Based Early Decision-Making (ML-EDM)


Green and Local Energy Exchanges and Optimizing Network Consumption


Orange 2021 Research Exhibition: let’s prepare for the future!


Measuring network usage for each mobile application


The Internet of Things (IoT) meeting the needs of trading ports


Improving the energy efficiency of connected objects