2025Activity report​​Project-TeamERMINE

3.1 Context‌​‌

Telecommunication networks are ubiquitous​​ in our daily life.​​​‌ The network ecosystem and‌ its infrastructure are increasingly‌​‌ complex with more and​​ more involved actors and​​​‌ technologies which have to‌ interoperate. This complexity is‌​‌ illustrated in Figure 1​​, where

Description of​​​‌ the various actors on‌ the ICT (for Information‌​‌ & Communications Technology) ecosystem​​

the basic network paradigm​​ with network operators simply​​​‌ sending content from/to end‌ users has moved to‌​‌ a topology with many​​ players having different roles.​​​‌ A complete telecommunications system‌ has to integrate traditional‌​‌ networks capabilities but also​​ all other innovative Internet​​​‌ services into a single‌ service platform. Selfishness of‌​‌ actors, heterogeneity of requests​​ and technologies, interoperability of​​​‌ services and automation of‌ decisions are major challenges‌​‌ to be tackled.

ERMINE​​ deals with network operations​​​‌ management and the required‌ associated monitoring and measurements‌​‌ and performance analysis. Operation​​ network management means for​​​‌ us performance-based management, fault‌ analysis, or the design‌​‌ of procedures and strategies​​ for resource provisioning and​​​‌ quality-of-service fulfillment. On this‌ management side, we aim‌​‌ at looking at the​​ whole topic of networks​​​‌ economics, studying the best‌ economic strategies of actors‌​‌ of the digital economy,​​ their interactions and the​​​‌ potential need for regulation.‌ We also want to‌​‌ address resource management through​​ artificial intelligence for automated​​​‌ decision as another main‌ research direction. On the‌​‌ measurement side, monitoring services​​​‌ of all sorts is​ a major challenge for​‌ regulatory bodies to ensure​​ a fair and legal​​​‌ behavior of providers, and​ is also of primary​‌ importance for providers themselves​​ to optimize decisions; this​​​‌ often requires new modeling,​ analysis and artificial intelligence​‌ tools. This is detailed​​ below.

3.2 Managing network​​​‌ operations

The heterogeneity and​ complexity of the ICT​‌ (for Information & Communications​​ Technology) world bring a​​​‌ number of challenges to​ network operations management. ERMINE​‌ addresses key issues for​​ network management: i) the​​​‌ challenges of ICT economics;​ ii) network control and​‌ interoperability iii) automated decisions​​ through AI.

ICT is​​​‌ omnipresent in our modern​ society, and the economy​‌ has gone beyond the​​ industrial economy to the​​​‌ Internet and ICT economy.​ Thanks to hyper-connectivity, there​‌ are now lots of​​ opportunities for innovation. As​​​‌ of May 2020, among​ the top-10 most valuable​‌ companies worldwide, seven are​​ ICT companies, trusting​​​‌ the places from 2​ to 8 (Microsoft, Apple​‌ Inc., Amazon Inc., Alphabet​​ Inc., Facebook, Alibaba Group,​​​‌ Tencent). Internet Service Providers​ (ISPs), Content Delivery Networks​‌ (CDNs) and cloud providers,​​ social network actors, all​​​‌ services and content providers​ are among actors needing​‌ a business model as​​ profitable as possible. Designing​​​‌ and analyzing such business​ models and their acceptance​‌ by end users are​​ issues to be addressed,​​​‌ leading to challenges in​ terms of outcome for​‌ all players (who act​​ in most cases selfishly)​​​‌ and in terms of​ benefits for society. On​‌ that last point, regulators​​ have to determine rules​​​‌ that actors need to​ follow in order to​‌ avoid harmful behaviors and​​ maximize social welfare. The​​​‌ issues we address include:​ the design of economic​‌ rules for new services,​​ the analysis of the​​​‌ impact of players decisions​ and interactions, and the​‌ potential design of rules​​ or incentives from regulatory​​​‌ bodies leading to the​ most adequate (social) outcome.​‌ One of the frameworks​​ to be used is​​​‌ that of game theory,​ and in particular of​‌ mechanism design.

There has​​ been a metamorphosis in​​​‌ the last few years​ on the management of​‌ network operations, driven by​​ the virtualization of networks​​​‌ and services. This evolution​ allows to meet the​‌ needs in terms of​​ the dynamic scaling of​​​‌ infrastructures and the agility​ of the decision-making, namely,​‌ a necessary prerequisite for​​ reducing operating costs as​​​‌ well as improving return​ on investment. These developments​‌ are radically changing the​​ way services are managed,​​​‌ as they become more​ complex (i.e., services in​‌ the form of graphs,​​ micro-services, etc.), their expectations​​​‌ can be very diverse​ and strategies for their​‌ placement could consider a​​ single domain or spanning​​​‌ across several domains, whether​ cooperative or not. Thus,​‌ the operation of the​​ network becomes extremely difficult​​​‌ and requires not only​ optimization of resources but​‌ also economic considerations, especially​​ when management involves several​​​‌ domains. We believe that​ automated management and control​‌ is the key direction​​ for an efficient solution.​​​‌ We deal with the​ automation of the network​‌ by contributing to ongoing​​ standardization efforts, notably by​​ the “Zero touch network​​​‌ & Service Management” (ZSM)‌ group of the ETSI‌​‌ (the European Telecommunications Standards​​ Institute), and through the​​​‌ elaboration of solutions based‌ on the most recent‌​‌ advances in machine learning​​ techniques, and in particular​​​‌ in deep reinforcement learning.‌ Some of the challenges‌​‌ we tackle include dynamic​​ placement of complex and​​​‌ constrained – QoS/QoE –‌ services (i.e., network slicing),‌​‌ automatic service scaling, congestion​​ control in the context​​​‌ of new generation networks,‌ including IoT networks (e.g.,‌​‌ massive Machine-Type Communications (mMTC)).​​ Different aspects are taken​​​‌ into account when developing‌ the various solutions, including‌​‌ reliability, resilience, guaranteed performance​​ (i.e., deterministic networks), but​​​‌ also the energy efficiency‌ on which the viability‌​‌ of the latter depends.​​ Ensuring the performance of​​​‌ the solutions when using‌ machine learning techniques is,‌​‌ however, a major problem​​ that is poorly addressed​​​‌ in the literature. We‌ address it by proposing‌​‌ techniques that offer at​​ least similar performances to​​​‌ heuristics. Indeed, even if‌ heuristics can be very‌​‌ efficient, they often have,​​ however, some limitations. One​​​‌ of these is the‌ blocking at the level‌​‌ of local minima. AI​​ techniques can address this​​​‌ issue encountered in combinatorial‌ problems (of the NP-hard‌​‌ type, such as the​​ travelling salesman one). They​​​‌ sometimes yield near-optimal results.‌ AI techniques have the‌​‌ additional advantage of being​​ able to learn by​​​‌ interacting with the environment,‌ and can therefore find‌​‌ solutions that heuristics could​​ not.

3.3 Measuring

A​​​‌ proper management of network‌ services and operations cannot‌​‌ be effective without measurement/monitoring​​ and without the analysis​​​‌ of relevant metrics. Indeed,‌ decision making cannot be‌​‌ realized without knowledge and​​ data on the past,​​​‌ present and even future‌ status of the activity.‌​‌ This is typically the​​ case for agile capacity​​​‌ planning and resource sharing‌ for which usage needs‌​‌ to be computed and​​ even predicted. This is​​​‌ a challenge for which‌ tools coming from AI‌​‌ are very promising. Another​​ relevant example of challenge​​​‌ is measurements for regulation‌ purposes, in relation with‌​‌ our activity on network​​ economics: defining regulation rules​​​‌ means designing measurement procedures‌ to verify that the‌​‌ rules are followed. It​​ is for example required​​​‌ to monitor a neutral‌ behavior of ISPs as‌​‌ expected from the Net​​ neutrality principles. While there​​​‌ exist a few tools‌ towards that goal, they‌​‌ are actually all devoted​​ to very specific hindrances​​​‌ to neutrality (blocking, degradation)‌ for specific types of‌​‌ flows or traffic; a​​ major challenge is to​​​‌ detect any type of‌ non-authorized behavior. Many other‌​‌ actors of the ICT​​ economy, if not all,​​​‌ are barely monitored while‌ an unfair behavior could‌​‌ be seen as harming​​ end users and society.​​​‌ To name some noticeable‌ examples, we can cite‌​‌ CDNs which cache some​​ content at the edge​​​‌ of the network and‌ could unfairly propose a‌​‌ better quality to selected​​ customers, or search engines​​​‌ who can prioritize the‌ web. In all those‌​‌ cases, different and specific​​ techniques have to be​​​‌ designed to analyze and‌ detect unexpected behaviors.

Similarly,‌​‌ managing resources and operations​​​‌ calls for evaluating (measuring)​ the impact of decisions​‌ to determine the most​​ appropriate ones. Modeling and​​​‌ performance evaluation techniques are​ appropriate and useful solutions​‌ at a low cost,​​ without the need to​​​‌ build and run the​ real system. While we​‌ have in the group​​ a long experience on​​​‌ performance evaluation, new challenges​ keep popping up due​‌ to new types of​​ services translated into new​​​‌ problems to be modeled​ and analyzed through performance​‌ evaluation tools, and adaptations​​ or extensions of existing​​​‌ techniques for better decision​ making. They usually require​‌ to develop new appropriate​​ network metrics to assess​​​‌ network service operations effectiveness.​ Illustrative examples are blockchains.​‌ The popularity of blockchain​​ lies in the introduction​​​‌ of the concept of​ a public distributed ledger​‌ shared by all participants​​ of the system without​​​‌ relying on a centralized​ authority. This distributed ledger​‌ records all the transactions​​ between parties efficiently and​​​‌ in a verifiable and​ permanent way. In order​‌ to enjoy higher throughput​​ and self-adaptivity to transactions​​​‌ demand in ICT, it​ requires the development of​‌ a new architecture and​​ thus specific modeling and​​​‌ analysis techniques.

3.4 Research​ axes

Our activity is​‌ organized in five subtopics​​ (or axes). The first​​​‌ two axes are on​ network management, and the​‌ next two are orthogonal​​ on the associated and​​​‌ required measurements. Axis 5​ on AI is transverse.​‌ The axes and their​​ intersections are described in​​​‌ Figure 2.

The​ five research axes: two​‌ on network management, two​​ orthogonal on measurements and​​​‌ transverse on AI, with​ intersections

As a​​ brief description:

• Network Economics.​​​‌ The digital economy has​ gained and keeps gaining​‌ in scale, scope, and​​ significance. The ecosystem is​​​‌ quickly evolving and one​ of our main goals​‌ is to answer all​​ questions related to the​​​‌ Internet & digital economy​ that pop up in​‌ line with what we​​ started to do, and​​​‌ to be reactive to​ the news in that​‌ domain. We want to​​ address issues concerning Internet​​​‌ resource allocation and pricing​ models, the economics of​‌ services, issues with vertical​​ integration, the economics of​​​‌ structuring platforms, as well​ as economics and regulation​‌ (including network neutrality).
• Managing​​ next generation networks &​​​‌ massive IoT communication. One​ of the fundamental challenges​‌ in the traffic management​​ of new and emerging​​​‌ networking activities is to​ describe, analyze and control​‌ heterogeneous resources and complex​​ services, under energy constraints.​​​‌ While AI is a​ main research direction to​‌ address this problem, there​​ exist many other possibilities.​​​‌ The main questions of​ interest include the automation​‌ of infrastructure management (a​​ typical example being network​​​‌ slicing) or the massive​ access in IoT networks.​‌
• Measuring, monitoring & regulation.​​ With the rapid evolution​​​‌ of networks, many needs​ are appearing for the​‌ design of measurement techniques​​ associated with new services,​​​‌ but also for regulators​ to monitor network's activity.​‌ Measurements are of two​​ possible categories: passive, monitoring​​ existing traffic, or active,​​​‌ i.e., based on injecting‌ traffic to investigate the‌​‌ network behavior. The goal​​ of this research direction​​​‌ is to develop practical‌ network operations measurement techniques,‌​‌ from the two different​​ points of view: operators​​​‌ (for a better management)‌ and regulators (for monitoring).‌​‌
• Measuring performance metrics based​​ on models. Measuring performances​​​‌ has to be done‌ not only to observe‌​‌ and monitor directly a​​ service like in previous​​​‌ subsection, but also at‌ different phases such as‌​‌ at conception or to​​ propose enhancements. Indeed, network​​​‌ services management and decision‌ making cannot be applied‌​‌ efficiently without a valid​​ and accurate evaluation of​​​‌ performance metrics through the‌ construction and analysis of‌​‌ models. We are carrying​​ out such evaluations.
• Artificial​​​‌ Intelligence (AI) in networking.‌ AI is a transverse‌​‌ research axis, since it​​ is used for both​​​‌ management and measuring. We‌ are not only users‌​‌ of the technology but​​ also interested in the​​​‌ methodology itself, that is,‌ we also develop new‌​‌ techniques, based on our​​ networking problems but usable​​​‌ outside. This concerns mainly‌ supervised learning, in particular‌​‌ time series forecasting using​​ this type of learning,​​​‌ and reinforcement learning. From‌ the point of view‌​‌ of the tools, our​​ skills and work are​​​‌ essentially related to Random‌ Neural Networks, Reservoir Computing‌​‌ and deep neural architectures​​ in Reinforcement Learning.

5.1​​​‌ Impact of research results​

Our research on network​‌ economics and particularly neutrality​​ issues aims to view​​​‌ how, and if, regulation​ can help to improve​‌ social welfare, among other​​ goals. Typical examples are​​​‌ how tools SNIDE and​ DemoWehe described later explaining​‌ how potential deviations from​​ rules by respectively search​​​‌ engines and Internet service​ providers can be detected.​‌ Our modeling works on​​ neutrality also allow to​​​‌ investigate the appropriate rules​ toward a benefice to​‌ society. This has led​​ to collaborations with specialists​​​‌ of human sciences in​ information networks (R. Badouard​‌ from Pantheon-Assas and I.​​ Lyubareva from IMT Atlantique)​​​‌ to tackle the socially​ sensitive issue of bias​‌ in information networks, in​​ particular from search engines.​​​‌

We also work on​ so-called green streaming to​‌ analyze the interactions and​​ strategies of various vendors​​​‌ in the market and​ provide valuable insights that​‌ can contribute to the​​ sustainable growth of the​​​‌ Content Delivery Network (CDN)​ industry. By understanding the​‌ implications of energy pricing​​ and its impact on​​​‌ the market dynamics, this​ research endeavors to offer​‌ potential solutions and recommendations​​ to address the challenges​​​‌ and foster a more​ sustainable and efficient CDN​‌ ecosystem. Similarly, we have​​ started to work on​​​‌ the societal, environmental, and​ economic impacts of digital​‌ infrastructures through a Cifre​​ PhD with Orange Labs.​​​‌

6.1 Awards

The paper​ entitled "Efficient Key Grouping​‌ for Near-optimal Load Balancing​​ in Stream Processing Systems"​​​‌ by E. Anceaume, Y.​ Busnel, L. Querzoni, N.​‌ Rivetti and B. Sericola,​​ which was presented at​​​‌ the 9th ACM International​ Conference on Distributed Event-Based​‌ Systems (DEBS'15), Oslo, Norway,​​ in July 2015, received​​ the "DEBS 10 Years​​​‌ Test of Time Award"‌ at the 19th ACM‌​‌ International Conference on Distributed​​ Event-Based Systems (DEBS’25), Gothenburg,​​​‌ Sweden, in June 2025.‌ This award honors research‌​‌ that has had a​​ lasting and significant impact​​​‌ on the field of‌ distributed event-based systems, even‌​‌ a decade after its​​ original publication. It highlights​​​‌ pioneering work that remains‌ influential and relevant over‌​‌ time.

7.1 Latest software developments‌

8.1 Network Economics

Participants:‌ Burak Kara, Maël‌​‌ Le Treust, Patrick​​ Maillé, Orland-Medy Saizonou​​​‌, Bruno Tuffin.‌

Network economics is a‌​‌ key topic to manage​​ and comprehend next generation​​​‌ networks, highlighted by the‌ edited book 36.‌​‌ From a high level​​ point of view, paper​​​‌ 47 also studies the‌ economic interactions between network‌​‌ stakeholders in next-generation mobile​​ networks. This paper presents​​​‌ a comprehensive analysis of‌ 5G and 6G ecosystems‌​‌ from an economic perspective,​​ including a classification of​​​‌ service classes, applications, and‌ enabling technologies. We map‌​‌ all relevant stakeholders, examine​​ their interactions, and discuss​​​‌ the business models and‌ pricing strategies yielding revenue.‌​‌ It also highlights the​​ regulatory frameworks helping to​​​‌ lead to expected societal‌ outcomes. The study exposes‌​‌ insights on sustainable business​​ strategies and policy development​​​‌ in next-generation communications.

But‌ we also work on‌​‌ specific topics.

Economics and​​ regulation: network neutrality and​​​‌ network slicing.

Network slicing,‌ a key feature of‌​‌ next-generation mobile networks, has​​ underexplored economic implications. We​​​‌ model in 28,‌ 48 a Stackelberg game‌​‌ where a mobile operator​​​‌ sells a slice to​ a service provider offering​‌ two service tiers, showing​​ that slicing boosts revenues​​​‌ and improves resource allocation.​

Economics of platforms.

Digital​‌ platforms shape users' choices​​ and opinions through the​​​‌ ranking of the displayed​ items, the top-ranked ones​‌ being more likely to​​ be considered than the​​​‌ next ones. While we​ expect the ranking to​‌ be based on some​​ sort of relevance, there​​​‌ is a suspicion that​ platforms could alter their​‌ ranking to promote some​​ content for financial or​​​‌ economic reasons, among other​ possibilities. We consider in​‌ 45 a model of​​ a ranking platform implementing​​​‌ revenue-maximizing rankings, and analyze​ the strategy of sellers​‌ paying to be ranked​​ higher for a better​​​‌ exposition, thereby increasing sales.​ We investigate the impact​‌ of such strategic behaviors​​ on revenues, overall click​​​‌ rates for all actors,​ and global relevance perceived​‌ by users.

Search neutrality.​​

The digital transformation has​​​‌ revolutionized information consumption, with​ search engines playing a​‌ pivotal role in shaping​​ user access to diverse​​​‌ media. Employing algorithms, these​ engines influence content visibility​‌ and aggregate news sources,​​ significantly molding public opinion.​​​‌ As gatekeepers of information,​ search engines impact media​‌ outlet visibility, affecting online​​ traffic, revenue, and journalistic​​​‌ diversity. In breaking news​ and societal issues, search​‌ engines expedite information dissemination,​​ influencing initial narratives. Focusing​​​‌ on movements against police​ violence, 1 conducts a​‌ comparative analysis across 12​​ search engines for terms​​​‌ "Black Lives Matter" and​ "Justice pour Adama". Our​‌ innovative methodology identifies biases​​ in information diversity, providing​​​‌ insights into the dynamics​ shaping visibility of societal​‌ issues.

Platforms regulation.

The​​ regulation of platforms has​​​‌ been a hot topic,​ particularly in Europe with​‌ the Digital Markets Act​​ (DMA) passed in 2022.​​​‌ The purpose of 22​ is to design a​‌ mathematical model representing a​​ game between long-term revenue-oriented​​​‌ platforms in competition playing​ with their ranking strategy​‌ of items. The objective​​ of platforms is to​​​‌ apply a trade-off between​ short-term revenue from each​‌ visit by displaying most​​ profitable items and long-term​​​‌ number of visits due​ to the satisfaction of​‌ users from the relevance​​ of the displayed items.​​​‌ We analyze the output​ of the game and​‌ the impact of proposed​​ regulation rules on platforms​​​‌ and users.

Similarly prevailing​ platforms may be forced​‌ by the DMA to​​ share with their competitors​​​‌ some, or all, of​ the data they collected​‌ on users that can​​ help improve service quality.​​​‌ Paper 44 proposes a​ model to analyze the​‌ impacts of such a​​ regulation, by comparing (i)​​​‌ a laisser-faire situation where​ prevailing platforms are free​‌ to sell data to​​ newcomer service providers, (ii)​​​‌ the case when platforms​ can decide to share​‌ data but charging newcomers​​ is prohibited, and (iii)​​​‌ a scenario where data​ sharing is imposed and​‌ enforced by the regulator.​​

Economics and energy reduction.​​​‌

The increasing focus on​ the environmental impact of​‌ video delivery is reshaping​​ the relationship between content​​​‌ providers (CPs), Content Delivery​ Network (CDN) vendors, and​‌ end-users. In 19,​​ we model the impact​​ of forthcoming energy regulations​​​‌ on the dynamics between‌ CPs, CDN vendors, and‌​‌ end-users.

Somewhat similarly, paper​​ 39 investigates how carbon​​​‌ emissions can be systematically‌ incorporated into telecommunication network‌​‌ capacity planning and pricing​​ decisions for network service​​​‌ providers to respond to‌ their drastic CO2 emission‌​‌ reduction objective. We introduce​​ a model in which​​​‌ user demand depends jointly‌ on price and congestion,‌​‌ while operators face both​​ monetary and environmental costs​​​‌ that scale with capacity‌ and served traffic. In‌​‌ detail, three regulatory scenarios​​ with their strategic interactions​​​‌ are examined through a‌ game theoretic lens. Our‌​‌ results show that environmental​​ constraints can substantially reduce​​​‌ carbon impact; They can‌ even increase social welfare,‌​‌ when environmental costs are​​ high, though often at​​​‌ the expense of operator's‌ revenue. The study offers‌​‌ quantitative insights into how​​ regulators and service providers​​​‌ can balance economic incentives‌ with climate-driven sustainability goals‌​‌ for future computer networks.​​

Information coding and decision​​​‌ making.

Papers 33 and‌ 50 investigate the vector‌​‌ version of the open​​ problem of Witsenhausen (1968).​​​‌ We extend the coding‌ scheme of Le Treust‌​‌ (2017) to the case​​ of a continuous alphabet,​​​‌ adapting the notion of‌ weak typicality. We determine‌​‌ new coding strategies and​​ evaluate their performance. Interestingly,​​​‌ we show that a‌ pair of auxiliary random‌​‌ variables, one discrete and​​ the other continuous, outperforms​​​‌ both the linear strategy‌ and that of the‌​‌ Witsenhausen counterexample (1968). By​​ judiciously choosing the auxiliary​​​‌ coding variables, we obtain‌ a novel compromise, better‌​‌ than previous solutions.

Paper​​ 6 investigate two unsolved​​​‌ zero-error problems: the source‌ coding problem with side‌​‌ information and the channel​​ coding problem. We focus​​​‌ our attention on families‌ of independent problems for‌​‌ which the probability distribution​​ decomposes into a product​​​‌ of probability distributions. A‌ crucial step is the‌​‌ additivity property of the​​ optimal rate, which does​​​‌ not always hold in‌ the zero-error regime, unlike‌​‌ in the vanishing error​​ regime. When the additivity​​​‌ holds, the concatenation of‌ optimal codes is optimal.‌​‌

8.2 Managing next generation​​ networks & massive IoT​​​‌ communication

Participants: Soraya Ait‌ Chellouche, Mouloud Atmani‌​‌, Yassine Hadjadj-Aoul,​​ Wilem Lamdani, Patrick​​​‌ Maillé, Fatima Zahra‌ Mardi, Gerardo Rubino‌​‌, Chetna Singhal.​​

Optimization of Massive-IoT emission​​​‌ strategies.

Long-Range Frequency-Hopping Spread-Spectrum‌ (LR-FHSS), the latest LoRaWAN‌​‌ physical layer standard, was​​ designed to enhance the​​​‌ uplink capacity of these‌ networks, a prerequisite for‌​‌ direct-to-satellite Internet of Things​​ deployments in areas lacking​​​‌ terrestrial infrastructures. As LR-FHSS‌ still employs an unslotted‌​‌ ALOHA channel-access scheme, its​​ network behavior can be​​​‌ accurately captured using the‌ classical ALOHA model. In‌​‌ 34, we introduce​​ the Multi-Class Multi-Channel ALOHA-based​​​‌ Model (MC2AMB), a novel‌ analytical model that accurately‌​‌ approximates packet delivery success​​ ratios and other metrics,​​​‌ enabling comprehensive evaluation of‌ LR-FHSS networks without requiring‌​‌ computationally expensive simulations. Simulation​​ results demonstrate that MC2AMB​​​‌ accurately estimates network performance‌ across various density scenarios,‌​‌ outperforming state-of-the-art models in​​ this regard. Furthermore, the​​​‌ model enables determining optimal‌ values for key parameters‌​‌ by leveraging its derivatives,​​​‌ providing valuable insights for​ network optimization.

Beyond the​‌ specific constraints of massive-IoT​​ connectivity, optimizing resource allocation​​​‌ remains a critical bottleneck​ across the entire wireless​‌ landscape. As networks evolve​​ to handle a surge​​​‌ in user equipment, enhancing​ spectral efficiency becomes paramount​‌ for higher throughput. Traditional​​ schedulers, such as Maximum​​​‌ Signal-to-Noise Ratio (MaxSNR), Proportional​ Fair (PF) perform well​‌ in specific contexts, but​​ can be outperformed in​​​‌ others. Moreover, there are​ many possible schedulers to​‌ try in order to​​ find the best ones​​​‌ for diverse contexts. An​ approach that uses Artificial​‌ Intelligence (AI) becomes a​​ conceivable solution. In 32​​​‌ we introduce a new​ AI tool that selects​‌ and evaluates schedulers that​​ enhance spectral efficiency for​​​‌ different contexts. Our approach​ is to train the​‌ AI model on several​​ traffic loads, to evaluate​​​‌ and select schedulers based​ on a general scheduling​‌ formula, resulting in an​​ explainable model for traffic​​​‌ scheduling. After the training,​ the appropriate scheduler is​‌ used for each traffic​​ load condition. In addition,​​​‌ our tool detects terms​ of the formula that​‌ do not have a​​ positive impact on enhancing​​​‌ spectral efficiency. Simulation results​ show that the obtained​‌ scheduler outperforms state-of-the-art algorithms​​ such as RR, MaxSNR​​​‌ and PF.

Conflict-Free Scheduling​ for Industrial IoT Networks.​‌

The Industrial Internet of​​ Things (IIoT) architecture is​​​‌ designed to minimize latency​ and transmission errors in​‌ industrial communication networks. The​​ IEEE 802.15.4e Time-Slotted Channel​​​‌ Hopping (TSCH) protocol was​ introduced to provide high​‌ reliability and robustness against​​ interference in such environments.​​​‌ However, collisions and external​ interferences in the native​‌ TSCH protocol can still​​ degrade network latency and​​​‌ other key performance metrics.​ To address these limitations,​‌ we work on proposing​​ a Conflict-Free Daisy Chain​​​‌ Scheduling (DCS)-TSCH, an enhanced​ and deterministic scheduling mechanism​‌ for IIoT networks. DCS-TSCH​​ is designed to avoid​​​‌ collisions and reduce interference,​ thereby improving network performance​‌ and consistency.

UAV-Assisted Mobile​​ Edge Computing for IoT​​​‌ Task Offloading.

The paper​ 2 investigates efficient task​‌ distribution for computation-intensive IoT​​ devices in urban environments​​​‌ where direct connectivity to​ edge servers is hindered​‌ by infrastructure barriers and​​ network congestion. We focus​​​‌ on multi-UAV-assisted architectures where​ UAVs act as communication​‌ relays collaborating with heterogeneous​​ edge servers to facilitate​​​‌ service delivery. A crucial​ strategy is the joint​‌ optimization of IoT-UAV-ES associations​​ and UAV network topology​​​‌ through preference-based stable matching​ and adaptive UAV deployment.​‌ When this approach is​​ properly optimized, the system​​​‌ effectively maximizes service provider​ profit while ensuring user​‌ satisfaction through stringent latency​​ guarantees and seamless connectivity​​​‌ to remote IoT devices.​

Optimization of EV charging​‌ strategies through reinforcement learning.​​

To maximize the use​​​‌ of renewable energy, and​ tackle its variability and​‌ uncertainty, flexible entities such​​ as electric vehicles can​​​‌ be used. The contributions​ 20, 21 present​‌ a scalable, decentralized multi-agent​​ system, where each electric​​​‌ vehicle (EV) seeks the​ best instants to charge​‌ to satisfy its mobility​​ needs, while favoring photovoltaic​​​‌ energy and avoiding congesting​ the electric network. Each​‌ EV makes autonomous decisions​​ using information from its​​ environment, using contextual multi-armed​​​‌ bandit algorithms based on‌ Thompson sampling.

8.3 Measuring,‌​‌ monitoring & regulation

Participants:​​ Yassine Hadjadj-Aoul, Burak​​​‌ Kara, Patrick Maillé‌, Gerardo Rubino,‌​‌ Bruno Tuffin.

Our​​ activity on monitoring has​​​‌ focused in 2025 on‌ specific contexts.

Network neutrality‌​‌ monitoring.

Network neutrality is​​ a sensitive topic worldwide.​​​‌ Laws are passed and‌ regulators are expected to‌​‌ monitor providers activities to​​ check if they conform​​​‌ to the enacted rules.‌ Therefore, tools need to‌​‌ be designed towards that​​ goal but require strong​​​‌ theoretical validations if used‌ for justifying legal actions.‌​‌ Our paper 43 reviews​​ the statistical grounds of​​​‌ the proposed tools, focusing‌ in particular on Wehe,‌​‌ co-designed by the French​​ regulator ARCEP. We provide​​​‌ a statistical analysis of‌ the method used in‌​‌ Wehe, compared with standard​​ ones.

Network slices monitoring.​​​‌

In the context of‌ the SmartNet DEFI collaboration‌​‌ between Nokia Bell Labs​​ and Inria, we work​​​‌ on proposing a novel‌ generalizable Network Tomography framework‌​‌ based on Graph Neural​​ Networks (GNNs) 38.​​​‌ Addressing the limitations of‌ classical algebraic and statistical‌​‌ inference methods, which are​​ often rigid and struggle​​​‌ with dynamic topologies ,‌ our approach leverages Relational‌​‌ Graph Convolutional Networks (RGCN)​​ and Line Graph transformations.​​​‌ This architecture ensures structural‌ invariance, enabling the model‌​‌ to learn from a​​ minimal set of monitors​​​‌ (e.g., just two) and‌ generalize to any other‌​‌ configuration without retraining. This​​ effectively achieves “zero-shot inference”,​​​‌ offering a robust and‌ scalable solution for real-time‌​‌ network observability.

Passive network​​ monitoring and troubleshooting from​​​‌ within the browser.

Despite‌ advances in network technologies,‌​‌ slow web browsing remains​​ a persistent issue, and​​​‌ its troubleshooting continues to‌ be challenging. To address‌​‌ this challenge, in 15​​, we presents a​​​‌ lightweight passive measurement solution‌ that relies on web‌​‌ performance data readily available​​ within the browser, such​​​‌ as the connect time,‌ the response time, and‌​‌ the page load time,​​ to infer network performance,​​​‌ detect anomalies, and troubleshoot‌ their origins. Through controlled‌​‌ network experiments with manually​​ injected anomalies, including multiple​​​‌ concurrent performance issues, and‌ leveraging a dataset of‌​‌ more than 43, 000​​ webpages and tens of​​​‌ thousands of network scenarios,‌ we develop a predictive‌​‌ model using machine learning​​ that is capable of​​​‌ estimating network performance metrics‌ with acceptable accuracy. By‌​‌ solely relying on users'​​ daily web activity, our​​​‌ solution can continuously monitor‌ network performance, identify anomalies,‌​‌ and provide actionable insights,​​ without overloading the network​​​‌ with measurement probes, thus‌ making network troubleshooting accessible‌​‌ and non-intrusive for everyday​​ users.

CDNs.

The content​​​‌ providers now leverage multiple‌ Content Delivery Networks (CDNs)‌​‌ to distribute the video​​ streams. Content steering is​​​‌ the main technology for‌ the implementation of smooth‌​‌ and dynamic multi-CDN orchestration.​​ 17 addresses the question:​​​‌ how to get informed‌ about the recovery of‌​‌ a faulty CDN if​​ no users are downloading​​​‌ from it?

Short-form videos‌ have emerged as a‌​‌ dominant force in online​​ media, they now account​​​‌ for a substantial share‌ of global media consumption.‌​‌ 18 analyzes a dataset​​​‌ of short-form videos from​ a popular news application,​‌ which includes metrics related​​ to videos consumed within​​​‌ the app in a​ format akin to YouTube​‌ Shorts or Instagram Reels.​​ Building on these insights,​​​‌ the paper poses open​ questions and advocates for​‌ community-driven efforts to establish​​ open protocols for short-form​​​‌ video content.

Performance of​ a Lightweight Consensus Cryptographic​‌ Protocol in WBANs. Wireless​​ Body Area Networks (WBANs)​​​‌ play a critical role​ in healthcare by enabling​‌ real-time monitoring through wearable​​ devices. Blockchain technology improves​​​‌ data integrity and security​ in WBANs, but introduces​‌ significant transmission and energy​​ challenges due to the​​​‌ constrained nature of these​ networks. 14 evaluates a​‌ lightweight consensus protocol that​​ uses Practical Byzantine Fault​​​‌ Tolerance (PBFT) principles, tailored​ to WBAN requirements, proposed​‌ by part of the​​ authors in previous work.​​​‌ Our study considers the​ main dynamics of the​‌ system, such as activation​​ and deactivation rates, as​​​‌ well as packet generation​ rates and transmission durations.​‌ Through packet-level system simulations,​​ the reliability and efficiency​​​‌ of the protocol are​ analyzed. This work provides​‌ actionable insights into optimizing​​ blockchain-based WBAN systems, highlighting​​​‌ the tradeoffs between node​ activation rates and network​‌ stability.

8.4 Measuring performance​​ metrics based on models​​​‌

Participants: Gerardo Rubino,​ Bruno Sericola, Bruno​‌ Tuffin.

Our work​​ on performance analysis in​​​‌ 2025 can be decomposed​ in simulation, with application​‌ to resilience, transient and​​ equilibrium analysis, with application​​​‌ to protocols, and distributed​ systems. It concerns both​‌ teoretical and practical problems.​​

Monte Carlo simulation.

We​​​‌ have continued our work​ on rare event simulation.​‌

Consider an estimand that​​ is a known smooth​​​‌ function of an unknown​ vector of parameters, such​‌ as means, probabilities, quantiles,​​ etc. Some but not​​​‌ necessarily all of the​ parameters relate to rare​‌ events, which typically make​​ them difficult to estimate​​​‌ with naive Monte Carlo,​ so variance-reduction techniques must​‌ to be applied to​​ obtain efficient estimators of​​​‌ those parameters. We develop​ in 46 conditions ensuring​‌ that the overall estimand​​ can be efficiently estimated,​​​‌ providing both sufficient and​ necessary and sufficient conditions​‌ on the parameter estimations​​ to lead to an​​​‌ efficient overall estimation. The​ paper illustrates the applicability​‌ of the theory through​​ many examples and numerical​​​‌ results.

Randomized Quasi-Monte Carlo.​

Consider again estimating a​‌ known smooth function (such​​ as a ratio) of​​​‌ unknown means. 23 accomplishes​ this by first estimating​‌ each mean via randomized​​ quasi-Monte Carlo and then​​​‌ evaluating the function at​ the estimated means. We​‌ prove that the resulting​​ plug-in estimator obeys a​​​‌ central limit theorem by​ first establishing a joint​‌ central limit theorem for​​ a triangular array of​​​‌ estimators of the vector​ of means and then​‌ employing the delta method.​​

Network reliability.

Network reliability​​​‌ computation is an NP-hard​ problem which has attracted​‌ much attention in literature.​​ Given the difficulty to​​​‌ compute the exact value​ of this metric, an​‌ alternative which has been​​ much explored in the​​​‌ literature is the use​ of Monte Carlo estimation​‌ methods. In 3,​​ we discuss Permutation Monte​​ Carlo, a highly efficient​​​‌ network reliability estimation method‌ but prone to numerical‌​‌ limitations for large networks.​​ We present a simple​​​‌ way to rewrite the‌ algorithm’s calculations to make‌​‌ it numerically more stable.​​ We also present a​​​‌ variant implementation for homogeneous‌ models (when all the‌​‌ links share the same​​ failure probability). For this​​​‌ type of network the‌ procedure can be redesigned‌​‌ to be much more​​ efficient. We present some​​​‌ experimental test results proving‌ that both proposed implementation‌​‌ variants perform extremely well.​​

Markov chains.

Sigmund duality​​​‌ makes a connection between‌ arbitrary irreducible Markov chains‌​‌ and specific absorbing ones.​​ In 11, we​​​‌ present a different way‌ of connecting both classes‌​‌ of chains, that proves​​ useful in the analysis​​​‌ of their steady-state regime‌ (for absorbing chains, instead‌​‌ of the useless steady-state​​ probabilities we work with​​​‌ absorption probabilities and with‌ mean absorption times). The‌​‌ connection we discuss also​​ considers arbitrary irreducible models,​​​‌ which are associated with‌ specific absorbing chains, but‌​‌ with a totally different​​ structure than when using​​​‌ duality.

12 deals with‌ the performance analysis of‌​‌ a system modeled by​​ a Markovian queue in​​​‌ its transient phase. Instead‌ of considering the classical‌​‌ transient distribution of the​​ model's state, we focus​​​‌ on the maximal number‌ of customers in the‌​‌ system, in the interval​​ $[0,\mathrm{t‌}]$. Since this‌ is a random variable,‌​‌ we evaluate its distribution.​​ The numerical scheme that​​​‌ we propose is based‌ on the Uniformization procedure,‌​‌ and we illustrate it​​ on the $M/‌M/1$ and‌ $M/M/‌‌1/H$ models.​​ We also show how​​​‌ to analyze another metric,‌ namely the fraction of‌​‌ lost customers in a​​ finite storage model, following​​​‌ a similar approach.

The‌ chapter 35 discusses two‌​‌ different but related problems:​​ the analysis of the​​​‌ regions of the models'‌ parameters where generalized steady-state‌​‌ distributions exist, for certain​​ real somewhat stochastic matrices​​​‌ $P$, and the‌ known relationship between the‌​‌ generalized steady-state distribution $\mathrm{\pi }$ of matrix $P$ and​​​‌ Gambler’s ruin probabilities defined‌ in the algebraic dual‌​‌ $P^{*}$ of $\mathrm{P}$, using our extension​​​‌ to Siegmund duality to‌ arbitrary matrices. The article‌​‌ is illustrated with many​​ examples of Markovian queueing​​​‌ systems.

Measuring the performance‌ of distributed systems.

Distributed‌​‌ systems protocols and network​​ protocols sometimes make use​​​‌ of random walks (i.e.,‌ token circulation) on such‌​‌ systems, to explore them​​ in order to collect​​​‌ information from them and‌ to make systems reliable.‌​‌ Important performance measures of​​ these protocols such as​​​‌ the search time and‌ the time to visit‌​‌ all the sites obviously​​ depend on some quality​​​‌ measures on random walks‌ such as the hitting‌​‌ and the cover times.​​ The cover time of​​​‌ a discrete-time homogeneous Markov‌ chain or of a‌​‌ random walk is the​​ time needed to visit​​​‌ all the states of‌ the process. We analyze‌​‌ in 4 the moments​​ and the distribution of​​​‌ hitting times on the‌ lollipop graph which is‌​‌ the graph exhibiting the​​​‌ maximum expected hitting time​ among all the graphs​‌ having the same number​​ of nodes. We obtain​​​‌ recurrence relations for the​ moments of all orders​‌ and we use these​​ relations to analyze the​​​‌ asymptotic behavior of the​ hitting time distribution when​‌ the number of nodes​​ tends to infinity. We​​​‌ consider in 5 the​ moments and the distribution​‌ of the hitting and​​ cover times of a​​​‌ random walk in the​ complete graph. We study​‌ both the time needed​​ to reach any subset​​​‌ of states and the​ time needed to visit​‌ all the states of​​ a subset at least​​​‌ once. As we did​ for the lollipop graph,​‌ we obtain recurrence relations​​ for the moments of​​​‌ all orders and we​ use these relations to​‌ analyze the asymptotic behavior​​ of the hitting and​​​‌ cover times distributions when​ the number of states​‌ tends to infinity. The​​ same type of results​​​‌ have been derived in​ 49 for the path​‌ graph and in 41​​ for both the star​​​‌ graph and the sun​ graph.

Continuing our collaboration​‌ with the Pirat Inria​​ Team, We consider in​​​‌ 40 a large scale​ and permissionless system, that​‌ is a system in​​ which users can join​​​‌ and leave at any​ time without any preliminary​‌ indication to their application.​​ We also consider that​​​‌ a percentage of these​ users are malicious, that​‌ is exhibit an arbitrary​​ behavior. These users can​​​‌ collide together. This is​ modeled by an oblivious​‌ adversary. In such a​​ system, we propose a​​​‌ probabilistic consensus algorithm that​ allows all the correct​‌ users to decide on​​ the same set of​​​‌ values. Specifically, with any​ probability larger than $\mathrm{1‌}-\epsilon$, if​​ any two correct users​​​‌ decide a set of​ values, then those sets​‌ are exactly the same.​​ This holds even if​​​‌ the adversary tries to​ prevent convergence to the​‌ same set of values.​​ Handling the adversarial nature​​​‌ of the system is​ achieved through proof-of-eligibility, ephemeral​‌ participation and the introduction​​ of a broadcast factor​​​‌ $\lambda$. Altogether, they​ guarantee with probability larger​‌ than $1-\mathrm{2}\epsilon$ that any transaction​​​‌ initially broadcast by any​ correct user will belong​‌ to the decision value,​​ and that any two​​​‌ decision values contain exactly​ the same set of​‌ transactions, whatever the adversarial​​ nature of the execution.​​​‌ Finally, any consensus execution​ terminates in six asynchronous​‌ rounds. A straightforward application​​ of our consensus algorithm​​​‌ is the design of​ permissionless distributed ledgers.

Our​‌ work on the comparison​​ of population protocols has​​​‌ led us to analyze​ the stochastic ordering of​‌ random variables. The stochastic​​ ordering is usefull to​​​‌ compare random variables in​ terms of their probability​‌ distribution: if $X$ and​​ $Y$ are two real​​​‌ random variables, we say​ $X$ is less than​‌ $Y$ in the usual​​ stochastic ordering whenever $\mathrm{P‌}(X>\mathrm{x})\le P(‌Y>x)$ for all $x\in ‌IR$. We​ then write $X{\le ‌}_{\mathrm{st}}Y$. For​​ instance if $X$ and​​ $Y$ are two durations​​​‌ then $X{\le }_{\mathrm{st‌}}Y$ means that for‌​‌ all values of $\mathrm{x}$, the duration $\mathrm{X‌}$ has less chances to‌ exceed $x$ than duration‌​‌ $Y$. In 42​​, we consider the​​​‌ case of sums of‌ independent and exponentially distributed‌​‌ random variables: if ${\mathrm{X}}_i$, resp. ${\mathrm{Y‌}}_i$ ($i=‌1,...,‌‌n$), are independent​​ random variables that are​​​‌ exponentially distributed with rates‌ ${\lambda }_i$, resp.‌​‌ ${\mu }_i$, we​​ examine under which conditions​​​‌ on the ${\lambda }_{\mathrm{i‌}}$'s and ${\mu }_{\mathrm{i‌‌}}$'s one can ensure​​ ${\sum }_{i=\mathrm{1‌}}^n{X}_i{\le ‌}_{\mathrm{st}}{\sum }_{i=‌‌1}^n{Y}_{\mathrm{i}}$.

Adopting a geomeric​​​‌ point of view in‌ the space of the‌​‌ parameters ${\lambda }_i$ and​​ ${\mu }_i$, we​​​‌ come up with close‌ to optimal conditions, improving‌​‌ previous results on this​​ question, as well as​​​‌ shedding a geometric light‌ on known results.

8.5‌​‌ Artificial Intelligence (AI) in​​ networking

Here, we are​​​‌ covering many different topics‌ in AI methods applied‌​‌ to networking problems, together​​ with some more methodological​​​‌ work.

Participants: Soraya Ait‌ Chellouche, Yassine Hadjadj-Aoul‌​‌, Gerardo Rubino,​​ Chetna Singhal, Lucas​​​‌ Taucoory, César Viho‌.

Green AI and‌​‌ Quantization for 6G.

With​​ the integration of AI​​​‌ as a native component‌ of 6G infrastructure, the‌​‌ energy consumption of ML​​ models becomes a critical​​​‌ concern. Paper 31 investigates‌ the impact of post-training‌​‌ quantization on LSTM, CNN,​​ and Transformer models for​​​‌ the predictive handover use‌ case. We benchmark quantized‌​‌ and full-precision models on​​ both CPU and GPU​​​‌ platforms to evaluate accuracy,‌ latency, and energy consumption.‌​‌ Our results show that​​ while quantization on CPUs​​​‌ delivers notable energy improvements,‌ particularly for CNN architectures‌​‌ and small batch sizes,​​ the gains on GPUs​​​‌ are inconsistent due to‌ limited INT8 support for‌​‌ sequence-based models. The study​​ provides practical guidance for​​​‌ deploying quantized models in‌ telecom environments, demonstrating that‌​‌ energy efficiency can be​​ improved without compromising predictive​​​‌ accuracy.

Orchestration of Dynamic‌ AI for Network Modeling.‌​‌

Paper 30 investigates the​​ problem of efficiently orchestrating​​​‌ Graph Neural Networks (GNNs)‌ for adaptive network modeling‌​‌ in mobile-edge-cloud continuum systems.​​ We focus our attention​​​‌ on heterogeneous network modeling‌ applications with diverse quality-of-service‌​‌ (QoS) and latency requirements​​ that must be met​​​‌ through dynamic configuration selection‌ and compute node assignment.‌​‌ A crucial step in​​ our approach is the​​​‌ representation of the system‌ using a tripartite graph‌​‌ model that captures the​​ relationships between applications, GNN​​​‌ model configurations, and heterogeneous‌ computing resources (CPUs, GPUs,‌​‌ and TPUs). When the​​ energy-efficient configuration selections are​​​‌ properly optimized through quantum‌ approximate optimization (QAO), the‌​‌ dynamic orchestration of training,​​ updating, and inference tasks​​​‌ yields near-optimal energy efficiency‌ while meeting the stringent‌​‌ application requirements. Our framework,​​ QAG, demonstrates that this​​​‌ adaptive orchestration strategy achieves‌ at least 50% energy‌​‌ savings and 60% lower​​ churn-rate compared to existing​​​‌ fixed-configuration approaches.

Network Traffic‌ Classification for Multi-Activity Situations‌​‌ Detection.

Classifying network traffic​​​‌ is one of the​ most crucial aspects of​‌ monitoring. The major contribution​​ of this work 37​​​‌ resides in the development​ of the SMAR (Single​‌ and Multi Activity Recognition)​​ framework. The main objective​​​‌ of the framework’s conception​ was to address the​‌ complex and real-life situation​​ of a user engaging​​​‌ in multiple digital activities​ simultaneously. In order to​‌ achieve this objective, we​​ first provided our proper​​​‌ conceptual and operational definitions​ of this concept (i.e.,​‌ multi-activity situations). Then, a​​ methodology was proposed for​​​‌ the generation of multi-activity​ network traces from single-activity​‌ traces, which was then​​ implemented. This methodological approach​​​‌ facilitated the establishment of​ a unique and essential​‌ dataset for the training​​ and validation of the​​​‌ proposed solution. The evaluation​ of the SMAR framework,​‌ based on deep learning​​ models and more specifically​​​‌ a multi-task learning architecture,​ has shown very encouraging​‌ performance. The outcomes of​​ this study not only​​​‌ confirmed the effectiveness of​ the proposed methodology in​‌ successfully differentiating between single-activity​​ and multi-activity scenarios, but​​​‌ also in accurately identifying​ the types of activities​‌ and applications involved.

Federated​​ Representation Learning for Application​​​‌ Type Detection in Future​ Mobile Networks.

Mobile​‌ application classification is essential​​ for advanced network management​​​‌ and application-based QoS policy​ enforcement in future, AI-enhanced,​‌ beyond 5G and 6G​​ mobile networks. We propose​​​‌ to use AI methods​ to categorize applications as​‌ functional types (e.g., Video,​​ Audio, Browsing) despite encryption​​​‌ and limited labeled data.​ This year, we tackle​‌ these challenges through unsupervised​​ representation learning, which maximizes​​​‌ the use of abundant​ unlabeled data in mobile​‌ networks. Due to the​​ distributed nature of beyond​​​‌ 5G and 6G networks,​ we use this method​‌ in federated learning scenarios​​ and compare it to​​​‌ the centralized ones. Our​ findings highlight that unsupervised​‌ learning improves model performance,​​ especially with scarce labeled​​​‌ data. Additionally, federated learning​ provides effective results as​‌ compared to centralized methods​​ 16.

Advanced AI​​​‌ for Sustainable and Proactive​ 6G Orchestration.

Realizing the​‌ vision of 6G requires​​ orchestrators that are not​​​‌ only intelligent but also​ energy-conscious and privacy-preserving across​‌ distributed domains. We currently​​ address these challenges through​​​‌ two complementary frameworks that​ leverage advanced Deep Reinforcement​‌ Learning (DRL).

First, to​​ enable proactive resource management​​​‌ without compromising domain privacy,​ we introduce the HERO​‌ framework 24, 51​​. This framework couples​​​‌ DRL with Latent Ordinary​ Differential Equations (Latent ODEs)​‌ to encode historical states​​ into privacy-preserving latent spaces.​​​‌ By exchanging these latent​ states and aggregating them​‌ via a gated attention​​ mechanism, distributed orchestrators achieve​​​‌ a holistic vision of​ the network's future, allowing​‌ them to anticipate inter-domain​​ traffic shifts effectively.

Second,​​​‌ addressing the specific challenge​ of sustainability during scaling​‌ operations, Paper 13 proposes​​ a distributed, energy-efficient placement​​​‌ framework. This approach utilizes​ a Graph Neural Network​‌ (GNN) to encode the​​ complex substrate topology and​​​‌ resource states, feeding this​ representation into a DRL​‌ agent, optimizing a refined​​ energy consumption model that​​​‌ accounts for both computational​ and network-level power usage.​‌

ML-based Network Slicing.

Beyond​​ standard AI approaches, metaheuristics​​ offer powerful tools for​​​‌ solving the NP-hard Virtual‌ Network Embedding (VNE) problem‌​‌ in 6G slicing. We​​ propose a robust optimization​​​‌ framework based on the‌ Greedy Randomized Adaptive Search‌​‌ Procedure (GRASP). In 25​​, we tackle the​​​‌ challenge of service reliability‌ by introducing a two-phase‌​‌ optimization process: an initial​​ resource-efficient placement followed by​​​‌ a reliability enhancement phase.‌ This method achieves significant‌​‌ improvements in placement reliability​​ compared to traditional heuristics.​​​‌ Complementing this, paper 26‌ addresses the issue of‌​‌ local optima in placement​​ algorithms. We propose an​​​‌ Enhanced Exploration GRASP that‌ expands the search space‌​‌ through a broader candidate​​ selection and a multi-level​​​‌ neighborhood search, yielding superior‌ success rates and revenue-to-cost‌​‌ ratios for complex service​​ requests.

Enhancing Autonomous Driving​​​‌ Navigation Using AI.

The‌ multimedia streaming can be‌​‌ challenging in resource constrained​​ dynamic environment 8.​​​‌ The paper 29 investigates‌ the problem of enabling‌​‌ safe autonomous navigation and​​ high-level autonomy tasks such​​​‌ as exploration and surveillance‌ for resource-constrained nano-sized platforms‌​‌ in partially-known environments. To​​ support the robust connectivity​​​‌ required for such tasks,‌ the paper 7 introduces‌​‌ a low-complexity modified pilot-based​​ channel estimation (MPCE) method​​​‌ for IEEE 802.22 systems‌ that achieves up to‌​‌ 78% reduction in runtime.​​ In 29, we​​​‌ focus our attention on‌ a pocket-size 30-gram Crazyflie‌​‌ 2.1 drone equipped with​​ an ultra-low-power greyscale camera​​​‌ and limited onboard computational‌ resources. A crucial strategy‌​‌ is the splitting of​​ the navigation task between​​​‌ edge and onboard computing,‌ wherein a deep learning-based‌​‌ object detector executes on​​ external hardware while the​​​‌ reactive planning algorithm runs‌ on the drone itself.‌​‌ When this split-computing approach​​ is properly optimized through​​​‌ integrated sensing, computing, and‌ communication (ISCC) paradigm, the‌​‌ autonomous navigation system achieves​​ a command rate of​​​‌ approximately 8 frames-per-second and‌ object detection performance of‌​‌ 60.8 COCO mean-average-precision. Field​​ experiments validate the feasibility​​​‌ of the framework with‌ successful autonomous flight at‌​‌ speeds up to 1​​ m/s, demonstrating safe obstacle​​​‌ avoidance and waypoint navigation‌ in unknown environments while‌​‌ maintaining real-time compatibility between​​ communication delays and perception​​​‌ requirements.

Dynamic AI Orchestration‌ for Energy-Efficient Inference in‌​‌ Distributed Network System.

The​​ paper 9 addresses dynamic​​​‌ deployment of gated neural‌ networks with context-aware sensor‌​‌ fusion across heterogeneous infrastructure.​​ It develops quantile-constrained inference​​​‌ optimization for joint orchestration‌ of data sources, DNN‌​‌ architectures, and resource allocation.​​ The paper 10 focuses​​​‌ on distributed execution of‌ DNNs equipped with early‌​‌ exits in mobile-edge-cloud systems.​​ It introduces feasible inference​​​‌ graphs to optimally determine‌ layer splitting and deployment‌​‌ locations. Together, these approaches​​ achieve adaptive, energy-efficient inference​​​‌ that satisfies application latency‌ and quality requirements. Both‌​‌ papers demonstrate dramatic energy​​ reductions through intelligent model​​​‌ splitting and context-aware configuration‌ selection, enabling efficient AI-driven‌​‌ applications across resource-constrained network​​ infrastructure.

Forecasting Seasonal Extreme​​​‌ Temperatures. In recent years,‌ great progress has been‌​‌ made in the field​​ of forecasting meteorological variables,​​​‌ but the progress is‌ much slower in predicting‌​‌ climatological metrics. In 27​​, we consider the​​​‌ maximum daily temperatures in‌ the short, medium, and‌​‌ long term. A common​​​‌ approach here is to​ frame the study as​‌ a temporal classification problem​​ with the classes “maximum​​​‌ temperature above normal, normal​ or below normal". From​‌ a practical point of​​ view, we created a​​​‌ large historical dataset (from​ 1981 to 2018) collecting​‌ information from weather stations​​ located in South America,​​​‌ plus exogenous data from​ the Pacific, Atlantic, and​‌ Indian Ocean basins. We​​ present the different aspects​​​‌ of this work, finally​ based on the use​‌ of the AutoGluonTS platform.​​ We obtain results similar​​​‌ than those of the​ litterature, but using a​‌ single and standard server​​ instead of huge hardware​​​‌ platforms.

9.1 Bilateral contracts with​​ industry

9.2​​ Bilateral Grants with Industry​​​‌

10.1​‌ International initiatives

10.2 International research​‌ visitors

10.3 National initiatives

10.4‌​‌ Regional initiatives

11.1​ Promoting scientific activities

11.2 Teaching - Supervision​​​‌ - Juries - Educational‌ and pedagogical outreach

12.1‌ Publications of the year‌​‌