A joint laboratory to strengthen the link between SPIE ICS and Télécom SudParis

Antoine Lavignotte, a teacher-researcher in networks at Télécom SudParis, previously held the position of business engineer at SPIE ICS. " My role was to understand customers' needs and propose a technical architecture to ensure the operational maintenance of their networks ", he explains. " This experience gave me a cross-disciplinary knowledge of network projects, not only from a technical point of view, but also from a financial and business point of view. This expertise, which was later supplemented by a thesis, led him to join the Réseaux et Services de Télécommunications (RST) department at Télécom SudParis in 2014.

A partnership initiated via a teaching and research chair

" With my teacher-researcher colleagues, we noticed that students were losing interest in the field of networks ", he recalls. " So we wanted to start from scratch and breathe new life into the field, an approach that led to the creation of a new, in-depth track. Wishing to involve industrial partners in this initiative, the team presented its project to five companies covering the entire value chain, starting with SPIE ICS, a subsidiary of the SPIE Group and a digital services company specializing in integration, facilities management and networks. Like the latter, Bouygues Telecom, CNS Communications, SNCF Réseau and Nokia were enthusiastic about joining the adventure. And so, in 2017, the Télécom SudParis "Future Networks for Tomorrow's Services" teaching chair was born, through the Mines-Télécom Foundation and in partnership with Mines Saint-Étienne and IMT-BS. This was the first work-study program for general engineering students.

A success, since three years later, the chair was renewed, with four of the five partners, only Nokia having withdrawn. Its scope was even broadened, with the addition of a research dimension to the renewed teaching component. Two theses were conducted over the next three years, on the security of fiber architectures and heterogeneous networks.

From an initial CIFRE thesis to a joint laboratory

At the same time, SPIE ICS has shown a desire to go further, by setting up a CIFRE thesis with Télécom SudParis, with the aim of reinforcing its market differentiation through research. The company had identified strong potential for innovation in part of its business. " To ensure that its customers' networks remain in operational condition, SPIE ICS often intervenes in the event of a breakdown or incident," notes Antoine Lavignotte. " However, this monitoring work is accompanied by a colossal amount of data, which until now has been little exploited. The idea was to use this data to identify incidents upstream, as part of a predictive maintenance approach.

Killian Murphy's CIFRE thesis aimed firstly to process and anonymize the 10 TB of data supplied by SPIE ICS. Next, several AI models were studied in order to determine which were most relevant to this type of temporal data. The results have been published in several scientific journals.

Buoyed by this success, the two partners decided to scale up and repeat this initial success. In November 2024, this desire gave rise to the joint laboratory Laboratoire de Transformation des Entreprises par l'Innovation et le Numérique (LATIN), with the support of the TSN Carnot institute. " Our aim is to bring together researchers from Télécom SudParis, but also from other schools at the Institut Mines-Télécom and the Institut Polytechnique de Paris, and SPIE ICS engineers around common topics," explains Antoine Lavignotte. " For the company, the aim is to use the results of the theses to develop innovative services for its customers. And for the researchers, this collaboration gives us the opportunity to work on concrete cases, based on real data, as close as possible to the needs of a major industrial player. "

AI for predictive network maintenance

LATIN's broad scope is reflected in the name of the joint laboratory. However, three initial major themes have been identified: the use of AI in networks, cybersecurity and reducing the environmental footprint of digital services. But this list could be extended to include new topics, depending on the issues encountered by SPIE ICS over the next three years, the planned duration of the joint laboratory.

The first of these follows on from the aforementioned CIFRE thesis on AI-enabled predictive maintenance. Its aim is to build on the work already carried out, and to work with SPIE ICS engineers to develop a tool that exploits these results and can be used in the field.

" This thesis also meets a need for interconnection," adds Antoine Lavignotte. "As an integrator, SPIE ICS calls on several manufacturers for its network equipment. However, each has its own supervision and even AI solution, in a closed environment, which can be a hindrance for the company and its customers. Our aim is to develop a system that is 100% adaptable, whatever the manufacturer of the equipment. This work will continue to be carried out by Killian Murphy, who now holds a doctorate from the Institut Polytechnique de Paris and has since been recruited by SPIE ICS to head up research within the company. He continues to maintain a strong link with the school, however, as he is also a visiting researcher at Télécom SudParis.

LLM applied to cybersecurity and application management

The second thesis focuses on cybersecurity and the contribution of a certain type of AI: generative AI. More specifically, it involves working on Large Language Models (LLMs). " In the event of an attack on a system, SPIE ICS' cybersecurity teams have to intervene as quickly as possible to close the loophole," says Antoine Lavignotte. " In such a situation, how could an LLM help an operator take the relevant measures as quickly as possible? We'll be looking into the best way to support the user in the field. "

Finally, the third thesis, already underway, still focuses on the benefits of LLMs, this time applied to the Kubernetes tool developed by Google. " This open source system enables us to deploy a service in a scalable way," explains Antoine Lavignotte. "For example, as soon as the number of users increases significantly, it ensures scalability by creating several clusters of servers in parallel, in order to absorb the flow and avoid overloading sites or excessive latency times. " The idea here would be to support SPIE ICS employees in their use of Kubernetes, via instructions tailored to each situation and provided in natural language, thanks to LLM.

Other thesis topics are currently being defined, notably on the theme of reducing the environmental footprint of digital services. The next step will be to identify the professors/researchers capable of supervising this work, the SPIE ICS teams concerned by the issue and available, and suitable candidates, starting by proposing to Télécom SudParis students interested in such an opportunity.

The LATIN team has set itself the target of ten theses by the end of 2027. " Nevertheless, this is a minimum," stresses Antoine Lavignotte. " We are not at all barred from going beyond that, depending on the needs identified by SPIE ICS and the problems encountered by its customers. "