Publications Scientifiques

Many real-world networks exhibit multi-scale organization, yet identifying nested structural patterns remains challenging. We introduce a recursive framework to uncover the hierarchical organization of networks by extracting and evaluating densely connected subgraphs, called local components. At each level, components are assessed based on modularity, stability, and minimum size before being further decomposed. We apply this method to 24 real-world networks from social, biological, and technological domains. The results show diverse hierarchical depths (from 1 to 5), with social networks typically exhibiting deeper and more complex structures. Our analysis highlights the relevance of mesoscopic decomposition for characterizing the internal architecture of networks and enables cross-domain comparisons of structural depth.

The Intelligent Reflecting Surface (IRS) is an enabling technology for beyond 5G and 6G. IRSs are passive reflective surfaces that can modify signals in a controlled manner. They are utilized to enhance coverage and spectrum with high energy efficiency and, therefore, are referred to as passive relays. This energy efficiency makes them a promising technology. The reflective nature of these surfaces allows for a line-of-sight path through the IRS between a transmitter and receiver when the direct path is obstructed with low-energy consumption. However, this requires careful positioning. Proper IRS positioning is crucial for optimizing network performance. Most studies consider a fixed placement of the IRS, but the IRS can also be mobile to better adapt to the dynamics of certain networks. In this paper, we propose to compare the fixed and mobile placements of several IRSs. The problem is modeled using an IRS activation strategy for each period, employing different comparison criteria.

The correlation of security alerts is crucial for effective threat detection and mitigation in modern cybersecurity landscapes. With the exponential growth in the volume of generated alerts, there arises an imperative to continually develop new, efficient methods for alert correlation. The use of Large Language Models (LLMs) presents new avenues for enhancing real-time threat detection in the field of cybersecurity. In this paper, we introduce MCM-Llama, a fine-tuned language model aimed at improving security event correlation for real-time threat detection. Building upon our prior research, we augment the current framework by integrating MCM-Llama and introducing a novel architecture to practically implement our proposition. In doing so, we replace the correlation module previously based on Named Entity Recognition (NER) and semantic similarity with a fine-tuned Large Language Model (LLM). This transition allows us to capitalize on the advanced capabilities of LLM in comprehending and correlating security events dynamically. Through this enhancement, we aim to significantly improve the accuracy and efficiency of our threat detection system while leveraging the full potential of LLM technology in cybersecurity. Our experimental results validate the efficacy of the proposed approach by significantly improving the correlation rate, thereby reinforcing the pivotal role of Large Language Models (LLMs) in advancing security event correlation and threat detection methodologies.

Modern Security Operations Centers (SOCs) struggle with an overwhelming influx of security alerts from diverse sources, leading to alert fatigue and delayed incident response times. Conventional Security Information and Event Management (SIEM) systems typically process alerts in isolation, missing complex attack patterns that unfold across multiple detection points. This research introduces a comprehensive machine learning framework for intelligent alert correlation using three classification approaches: binary correlation detection, correlation type classification, and incident classification. We evaluated eight machine learning algorithms—Logistic Regression, Random Forest, XGBoost, LightGBM, Artificial Neural Networks (ANN), Convolutional Neural Networks (CNN), CatBoost, and Support Vector Machines (SVM)—using 50,000 realistic security alerts generated from 2.30 million network flows in the CIC-IDS-2017 dataset. Our multi-classification framework incorporates temporal relationships, network topology, attack progression patterns, and severity indicators to enhance correlation precision. The experimental results reveal remarkable performance across all classification tasks. CatBoost achieved the highest overall performance with 99.27% average accuracy across all tasks, closely followed by Random Forest (99.17%) and LightGBM (99.06%). For multiclass correlation type classification, six out of eight models achieved perfect 100% accuracy, with only XGBoost and SVM showing decreases to 96.30%. Binary correlation detection demonstrated exceptional performance with three models (Random Forest, LightGBM, CatBoost) achieving 99.96% accuracy. The proposed framework demonstrates exceptional capability in reducing alert noise while maintaining optimal detection rates, significantly enhancing SOC operational efficiency and enabling more targeted threat response strategies.

The intelligent reflective surface (IRS) is a new technology that can reduce the number of base stations and improve signal quality. Whereas in the literature, IRS is often considered in one-hop scenarios, we propose to address two-hop IRS in a vehicular context. On the one hand, two-hop IRS should improve connectivity by enabling blind devices in blind areas to be connected to the network, and on the other, connections may be more volatile due to a higher number of intermediate devices. In this work, their impact on network performance is investigated through different performance criteria. Our approach is validated using real vehicular traces in the city of Roma, Italy.

Predicting the next phases of an ongoing cyberattack is critical for proactive cybersecurity, enabling defenders to anticipate and mitigate threats effectively. This paper presents a context-aware meta-model framework designed to dynamically select the most suitable prediction model for projecting the next phase of an attack. Depending on real-time situational parameters, the meta-model determines whether to employ a Graph Attention Network (GAT) or a Deep Q-Network (DQN). The GAT model captures complex, graph-based relationships in attack patterns, while the DQN model excels in sequential decision-making for evolving attack scenarios. Experimental evaluations using the CTU-13 dataset demonstrate the effectiveness of this approach: GAT achieves 99.20% accuracy, and DQN achieves 100% in predicting the second phase of an attack based on an alert set representing the first phase. These results …

In many domains, including social network analysis, biology, and beyond, community structure discovery in networks is a basic topic. To find highly connected node groups—often referred to as communities—in these networks, researchers develop a wide range of techniques. While the field evolves rapidly, many comparative studies are carried out to assess the strengths and weaknesses of different algorithms on various types of networks and scenarios. This study attempts to provide a thorough comparative analysis involving both established techniques like Louvain, Label Propagation, Infomap, Girvan-Newman, and Walktrap, as well as newly proposed methods like PercoMVC, Walkscan, and Paris. This is due to the constantly changing landscape of community detection algorithms and the critical importance of understanding their strengths and weaknesses. We carry out a thorough empirical assessment of these eight community recognition methods on a variety of real-world networks, such as biological, social, and collaborative networks, among others with different topologies. A wide range of internal and external quality criteria, including conductance, density-based measurements, modularity, cut ratio, and the Friedman rank test, are incorporated into our analysis. Louvain consistently performs well on dense networks and is the best at optimizing modularity. Large, sparse networks work best with label propagation, effectively reducing the number of intercommunity links. PercoMVC shows high overall resilience and adaptability across a variety of network configurations. Infomap is good at capturing complex community structures and works well on various network types. Walktrap provides a solid balance between accuracy and efficiency and demonstrated high performance on congested networks. Our study provides insights to guide practitioners in selecting the most suitable community detection technique based on the network characteristics at hand, highlighting the complementary strengths of algorithms like Louvain and Label Propagation on dense and sparse networks respectively, and PercoMVC’s versatility.

Network analysis is a crucial tool for understanding complex systems, offering insights into patterns, optimization, and behavior prediction across various fields. Traditional mesoscopic representations like community and core-periphery structures often face ambiguities. We propose a novel component structure framework that distinguishes between dense local components and sparse global components, offering clear definitions, compatibility with standard network analysis techniques, and the ability to integrate hierarchical relationships for a more nuanced and comprehensive network analysis.

Assessing strategies for managing dengue is crucial. While many studies have examined interventions' impact on transmission dynamics, few focus on combining multiple therapeutic approaches. This study presents an epidemic model incorporating vaccination and treatment for infected individuals. It employs ordinary differential equations with seven-state variables, analyzing disease-free and endemic equilibria. Results show the disease-free equilibrium is globally stable when the basic reproduction number is below one. When vaccine effectiveness is low, treatment proves more effective in reducing dengue cases. However, a highly effective vaccine alone significantly reduces dengue occurrences. Additionally, the study introduces an optimal control problem, combining vaccination and treatment. The analysis suggests implementing both strategies is more effective than a single approach in mitigating disease spread.

This paper introduces a component-aware ranking strategy to identify critical nodes in modular networks. It overcomes the limitations of traditional centrality-based methods that fail to account for mesoscopic complex network structuration. The component structure decomposes networks into local (densely connected areas) and global (bridges between these areas) components allowing for the flexible use of different centrality measures in each component to better identify key nodes. We evaluate the method using the Susceptible-Infected-Recovered (SIR) model on LFR synthetic networks that simulate varying levels of community strength, ranging from tightly knit to highly mixed communities. The experimental results demonstrate that the community-aware strategy significantly improves diffusion containment in networks with strong community structures compared to classical centrality ranking methods. Its performance is less pronounced in moderately modular networks, and in weakly modular networks, its benefits are minimal. Additionally, we explore the impact of combining centrality measures (e.g., Betweenness for global and Degree for local components), finding marginal improvements in moderately mixed networks but no substantial advantages in highly mixed ones. Our findings suggest that the component-aware strategy is particularly suited for networks with pronounced community structures, offering a robust tool for information-spreading applications where targeting influential nodes is crucial.

This paper presents a mathematical model to explore the interplay between vaccination and treatment strategies to control dengue fever, a significant public health concern in tropical and subtropical regions. The study employs an SVITR compartmental model (Susceptible, Vaccination, Infectious, Treatment, Recovered) to analyze dengue transmission dynamics and assess the effectiveness of various intervention strategies. Key findings indicate that while a highly effective vaccine can significantly reduce dengue cases, treatment becomes more critical when the vaccine's efficacy is low. The results show the advantages of combining vaccination and treatment to optimize disease control, providing valuable information for public health policy.

Identifying influential nodes in networks is a crucial task with many applications across various domains. Traditional centrality measures, while insightful, often fail to capture the true influence of nodes, especially in complex networks. Machine learning techniques can potentially incorporate diverse node features, but their effectiveness relies heavily on feature engineering. In this study, we propose a hybrid methodology that synergistically combines well-established centrality measures as topological features and employs a powerful Random Forest classifier. Our approach extracts degree, betweenness, closeness, eigenvector centrality, PageRank, and clustering coefficients for each node, which are then used as input features to the Random Forest model. We evaluate our method on three real-world networks: the Cora dataset, the CA-HepTh dataset, and the Facebook dataset. The results demonstrate the …

This study leverages game theory to examine attack and defense strategies in the global air transportation network. We focus on zero-sum games where attackers and defenders use centrality-based strategies. The research highlights the vulnerability of networks to different attack strategies and the corresponding defense mechanisms that can mitigate these attacks. Results show that when the attack and defense resources are equal, no single defense strategy proves uniformly effective, but strategies based on betweenness centrality show significant resilience. Analyzing Nash equilibrium in the global air network reveals diverse equilibrium strategies, emphasizing the need for varied defense approaches as attack ranges increase.

Evaluating public health intervention techniques is essential to manage dengue successfully. This study proposes an epidemic model that includes two essential therapeutic interventions: vaccination and treating infected individuals to better understand dengue transmission dynamics. We introduce a seven-state model and investigate the model's endemic and disease-free equilibria. Results show that when vaccination efficacy is low, therapy is a more effective strategy for lowering dengue cases, whereas a highly effective vaccine significantly reduces dengue cases. The study also formulates an optimal control problem, suggesting that combined strategies are more effective in mitigating the disease spread.

We introduce the "component structure," a mesoscopic representation that differentiates between a network's dense and sparse global components. Unlike its alternatives (community structure, core-periphery), the representation offers a straightforward definition, captures a broader spectrum of mesoscopic features, and supports hierarchical analysis, allowing for a multilevel network examination of networks' organizational principles and functional behaviors.

The proliferation of IoT is evident in numerous domains today. Its applications are expanding rapidly, offering significant benefits. Nonetheless, it is plagued by numerous security flaws that can hinder its adoption in critical areas. This paper addresses the development of an intrusion detection system (IDS) for IoT networks. Specifically, we implement an autoencoder, an unsupervised deep learning model frequently employed in anomaly detection. The model is trained using the MQTTSet dataset, which contains both normal MQTT traffic and attack data. Training focused exclusively on legitimate data. Testing was conducted on both benign and malicious data to assess the model's effectiveness. The results indicate detection rates of 99.86% and 98.56% for normal and attack data, respectively.

This study explores the robustness of the world air transportation network by analyzing its component structure. The robustness analysis assesses the impact of targeted attacks on key airports based on three centrality measures: strength, betweenness, and PageRank. Our findings reveal that removing airports with high betweenness centrality is the most effective strategy for disrupting the network, isolating regional components from the global network.

Dengue fever is a major health concern in tropical and subtropical regions, affecting millions annually. This study develops an epidemic model to evaluate the combined effects of vaccination and treatment on dengue transmission. The model uses ordinary differential equations to examine the dynamics of susceptible, vaccinated, infected, treated and recovered individuals in human and mosquito populations. Key findings indicate that a highly effective vaccine can significantly reduce dengue incidence, but optimal control is achieved through a combined approach of vaccination and treatment.

This study explores the strategic dynamics within interconnected systems by integrating game theory with complex networks. It presents a static zero-sum game model to analyze attack and defense strategies in such networks. Investigating three strategies for attackers and defenders—random, degree centrality, and betweenness centrality—the study examines Nash equilibrium under equal resource assumptions. Analyzing the payoff matrix and players’ responses identify the dominant strategy as combining random attacks and betweenness-based defenses.