Agentic AI in the dark : AI-driven offense and defense against attacks originating from local TOR networks / Roumaissa Imane Nakhlati

Public ISBD Titre : Agentic AI in the dark : AI-driven offense and defense against attacks originating from local TOR networks Type de document : document multimédia Auteurs : Roumaissa Imane Nakhlati, Auteur ; Messaoud Babaghayou, Directeur de thèse Editeur : Laghouat : Université Amar Telidji - Département d'informatique Année de publication : 2026 Importance : 111 p. Accompagnement : 1 disque optique numérique (CD-ROM) Note générale : Option : Networks, systems and distributed applications Langues : Anglais (eng) Mots-clés : Cybersecurity  Autonomous agents  Docker  TOR anonymization  Intrusion detection  Network security  Data sovereignty  AI-driven agents Résumé : DockSec AI is a fully local, autonomous cybersecurity framework for offensive and defensive assessments within a controlled TOR-anonymized laboratory environment. The system deploys two specialized AI agents—a Red Agent (offensive) and a Blue Agent (defensive)—that share a locally hosted language model to execute adaptive attack strategies and perform AI-driven threat detection. The Red Agent autonomously performs reconnaissance, exploit selection, credential harvesting, and post-exploitation through a private Docker-based TOR network, ensuring complete traffic anonymization. The Blue Agent monitors network traffic, classifies alerts using AI reasoning, applies TOR-aware risk scoring, and maintains persistent memory through vector-based storage. Experimental results demonstrate the framework’s effectiveness : the Red Agent successfully executes adaptive attacks with credential auto-trigger and replanning, while the Blue Agent achieves significant detection rate improvements over rule-based expert baselines while reducing false positives. All operations run locally on standard hardware using a small quantized model, proving that effective AI-driven security is accessible without cloud dependency or specialized infrastructure. DockSec AI ensures data sovereignty, privacy protection, and compliance through on-site Docker containerization, enabling rapid deployment, portability, and cost-efficient scalability for security research and autonomous assessments. note de thèses : Mémoire de master en informatique Agentic AI in the dark : AI-driven offense and defense against attacks originating from local TOR networks [document multimédia] / Roumaissa Imane Nakhlati, Auteur ; Messaoud Babaghayou, Directeur de thèse . - Laghouat : Université Amar Telidji - Département d'informatique, 2026 . - 111 p. + 1 disque optique numérique (CD-ROM). Option : Networks, systems and distributed applications Langues : Anglais (eng) Mots-clés : Cybersecurity  Autonomous agents  Docker  TOR anonymization  Intrusion detection  Network security  Data sovereignty  AI-driven agents Résumé : DockSec AI is a fully local, autonomous cybersecurity framework for offensive and defensive assessments within a controlled TOR-anonymized laboratory environment. The system deploys two specialized AI agents—a Red Agent (offensive) and a Blue Agent (defensive)—that share a locally hosted language model to execute adaptive attack strategies and perform AI-driven threat detection. The Red Agent autonomously performs reconnaissance, exploit selection, credential harvesting, and post-exploitation through a private Docker-based TOR network, ensuring complete traffic anonymization. The Blue Agent monitors network traffic, classifies alerts using AI reasoning, applies TOR-aware risk scoring, and maintains persistent memory through vector-based storage. Experimental results demonstrate the framework’s effectiveness : the Red Agent successfully executes adaptive attacks with credential auto-trigger and replanning, while the Blue Agent achieves significant detection rate improvements over rule-based expert baselines while reducing false positives. All operations run locally on standard hardware using a small quantized model, proving that effective AI-driven security is accessible without cloud dependency or specialized infrastructure. DockSec AI ensures data sovereignty, privacy protection, and compliance through on-site Docker containerization, enabling rapid deployment, portability, and cost-efficient scalability for security research and autonomous assessments. note de thèses : Mémoire de master en informatique

BPS-R : Behavioral and privacy-aware secure routing protocol for DTNs / Abir Baroud

Public ISBD Titre : BPS-R : Behavioral and privacy-aware secure routing protocol for DTNs Type de document : document multimédia Auteurs : Abir Baroud, Auteur ; Soumia Ikram Rais, Auteur ; Messaoud Babaghayou, Directeur de thèse Editeur : Laghouat : Université Amar Telidji - Département d'informatique Année de publication : 2026 Importance : 82 p. Accompagnement : 1 disque optique numérique (CD-ROM) Note générale : Option : Networks, systems and distributed applications Langues : Anglais (eng) Mots-clés : Delay-Tolerant Networks  Opportunistic routing  PRoPHET  Privacy-preserving routing  Pseudonym rotation  Location Obfuscation  Trust management Résumé : Delay-Tolerant Networks (DTNs) enable communication in intermittently connected environments through a store-carry-forward paradigm. Existing protocols such as PRoPHET, Spray and Wait, and Epidemic routing improve message delivery under disrupted connectivity; however, they commonly rely on persistent MAC addresses during packet exchanges. This exposes nodes to privacy risks, as passive observers can correlate transmissions over time and reconstruct user mobility patterns. This thesis presents Behavioral and Privacy-aware Secure Routing (BPS-R), a novel DTN routing protocol that integrates privacy protection mechanisms directly into forwarding decisions. BPS-R extends PRoPHET with a behavioral forwarding engine based on contact recency and regularity, an epoch-based pseudonym rotation mechanism, coarse-zone location obfuscation, and a delivery transmission range mechanism based on trust tiers. BPS-R was implemented in the OPS simulator on OMNeT++. Its performance was evaluated using 500 nodes under the SWIM mobility model across simulation scenarios of 6, 24, 48, and 96 hours, respectively. The results show that BPS-R achieves delivery ratios comparable to PRoPHET, reaching 33.59% at 96 hours, while reducing total communication traffic by approximately 86% relative to PRoPHET (45MB vs. 326.9MB at 96 hours). Against a global passive timing-correlation attacker, BPS-R provides strong identity privacy at the standard 900-second data generation interval, with an attacker success rate of only 4.2% (privacy score of 0.958) and an average candidate set size of 6.40 nodes per unresolved pseudonym, corresponding to meaningful k-anonymity. These findings show that privacy preservation and routing efficiency do not conflict significantly as objectives in DTN design. When privacy mechanisms are integrated into the routing architecture, PRoPHET-level delivery performance can be preserved while substantially reducing communication overhead and improving identity privacy. note de thèses : Mémoire de master en informatique BPS-R : Behavioral and privacy-aware secure routing protocol for DTNs [document multimédia] / Abir Baroud, Auteur ; Soumia Ikram Rais, Auteur ; Messaoud Babaghayou, Directeur de thèse . - Laghouat : Université Amar Telidji - Département d'informatique, 2026 . - 82 p. + 1 disque optique numérique (CD-ROM). Option : Networks, systems and distributed applications Langues : Anglais (eng) Mots-clés : Delay-Tolerant Networks  Opportunistic routing  PRoPHET  Privacy-preserving routing  Pseudonym rotation  Location Obfuscation  Trust management Résumé : Delay-Tolerant Networks (DTNs) enable communication in intermittently connected environments through a store-carry-forward paradigm. Existing protocols such as PRoPHET, Spray and Wait, and Epidemic routing improve message delivery under disrupted connectivity; however, they commonly rely on persistent MAC addresses during packet exchanges. This exposes nodes to privacy risks, as passive observers can correlate transmissions over time and reconstruct user mobility patterns. This thesis presents Behavioral and Privacy-aware Secure Routing (BPS-R), a novel DTN routing protocol that integrates privacy protection mechanisms directly into forwarding decisions. BPS-R extends PRoPHET with a behavioral forwarding engine based on contact recency and regularity, an epoch-based pseudonym rotation mechanism, coarse-zone location obfuscation, and a delivery transmission range mechanism based on trust tiers. BPS-R was implemented in the OPS simulator on OMNeT++. Its performance was evaluated using 500 nodes under the SWIM mobility model across simulation scenarios of 6, 24, 48, and 96 hours, respectively. The results show that BPS-R achieves delivery ratios comparable to PRoPHET, reaching 33.59% at 96 hours, while reducing total communication traffic by approximately 86% relative to PRoPHET (45MB vs. 326.9MB at 96 hours). Against a global passive timing-correlation attacker, BPS-R provides strong identity privacy at the standard 900-second data generation interval, with an attacker success rate of only 4.2% (privacy score of 0.958) and an average candidate set size of 6.40 nodes per unresolved pseudonym, corresponding to meaningful k-anonymity. These findings show that privacy preservation and routing efficiency do not conflict significantly as objectives in DTN design. When privacy mechanisms are integrated into the routing architecture, PRoPHET-level delivery performance can be preserved while substantially reducing communication overhead and improving identity privacy. note de thèses : Mémoire de master en informatique

End-to-End latency reduction techniques for delay-sensitive applications in satellite-based edge computing networks / Fatima Zahra Zaoui

Public ISBD Titre : End-to-End latency reduction techniques for delay-sensitive applications in satellite-based edge computing networks Type de document : document multimédia Auteurs : Fatima Zahra Zaoui, Auteur ; Messaoud Babaghayou, Directeur de thèse Editeur : Laghouat : Université Amar Telidji - Département d'informatique Année de publication : 2025 Importance : 67 p. Accompagnement : CD ROM Note générale : Option : Networks,systems and distributed applications Langues : Anglais (eng) Mots-clés : Satellite  Edge Computing  Task Offloading  End-to-End Latency  Mist Layer  Intelligent orchestration  Delay-sensitive applications Résumé : With the advent of satellite-edge convergence, which is transforming distributed computing to make services real-time and intelligent, the management of task offloading more efficiently is more important than before.This work addresses the issue of minimizing the end-to-end delay for delay-critical applications in satellite-fueled mist computing systems. This work formulates a new task orchestration algorithm named IsoLink, which was specifically developed and tuned for mist-layer offloading and optimized to classify tasks based on context and assign them wisely to the appropriate virtual machines.IsoLink functions by normalizing the delay and execution parameters, taking into account the type and resource proximity to the task, to enable optimal deployment. By simulating comprehensively using the SatEdgeSim framework, the algorithm performed well in terms of reduced latency, energy savings, and enhanced task success ratio. The evaluation was conducted across various performance metrics and also compared with algorithms such as the Weighte_Greedy, Round_Robin, and Random_VM, and proved the efficacy of IsoLink. Although the method is well-suited for services requiring low latency, future improvements make satellite mist computing environments more adaptive and efficient in operation. note de thèses : Mémoire de master en informatique End-to-End latency reduction techniques for delay-sensitive applications in satellite-based edge computing networks [document multimédia] / Fatima Zahra Zaoui, Auteur ; Messaoud Babaghayou, Directeur de thèse . - Laghouat : Université Amar Telidji - Département d'informatique, 2025 . - 67 p. + CD ROM. Option : Networks,systems and distributed applications Langues : Anglais (eng) Mots-clés : Satellite  Edge Computing  Task Offloading  End-to-End Latency  Mist Layer  Intelligent orchestration  Delay-sensitive applications Résumé : With the advent of satellite-edge convergence, which is transforming distributed computing to make services real-time and intelligent, the management of task offloading more efficiently is more important than before.This work addresses the issue of minimizing the end-to-end delay for delay-critical applications in satellite-fueled mist computing systems. This work formulates a new task orchestration algorithm named IsoLink, which was specifically developed and tuned for mist-layer offloading and optimized to classify tasks based on context and assign them wisely to the appropriate virtual machines.IsoLink functions by normalizing the delay and execution parameters, taking into account the type and resource proximity to the task, to enable optimal deployment. By simulating comprehensively using the SatEdgeSim framework, the algorithm performed well in terms of reduced latency, energy savings, and enhanced task success ratio. The evaluation was conducted across various performance metrics and also compared with algorithms such as the Weighte_Greedy, Round_Robin, and Random_VM, and proved the efficacy of IsoLink. Although the method is well-suited for services requiring low latency, future improvements make satellite mist computing environments more adaptive and efficient in operation. note de thèses : Mémoire de master en informatique

Réservation

Réserver ce document

Exemplaires (1)

Code-barres	Cote	Support	Localisation	Section	Disponibilité
MF 01-89	MF 01-89	CD	BIBLIOTHEQUE DE FACULTE DES SCIENCES	théses (sci)	Disponible

Machine learning for link prediction in complex networks / Messaoud Babaghayou

Public ISBD Titre : Machine learning for link prediction in complex networks Type de document : texte manuscrit Auteurs : Messaoud Babaghayou, Auteur ; Abdallah Lakhdari, Directeur de thèse Editeur : Laghouat : Université Amar Telidji - Département d'informatique Année de publication : 2016 Importance : 79 p. Format : 30 cm. Accompagnement : 1 disque optique numérique Note générale : Option : Networks, systems and distributed applications ( Réseaux,systèmes et applications réparties) Langues : Anglais (eng) Mots-clés : Machine Learning  Link Prediction  Complex Networks  Supervised Leaning  Unsupervised Learning  Classification  Node-based Metrics Résumé : Nowdays, networks are omnipresent. The study and understanding of these networks become a greater need. The purpose of this work, is to investigate link prediction task in complex networks using Machine learning techniques. In fact, we propose two approaches to perform link prediction: supervised and unsupervised one. In both techniques a link or a pair of nodes is characterized by several features based on network topology-based metrics. In addition, we investigate many combined features. Concerning the supervised approach, we investigate the KNN and decision tree methods to build the link prediction models. While in the unsupervised approach, we rely on ranking strategy. An experimental study is performed on real networks. The results show that the supervised approach using gathered features reaches good performances with 84% f-measure. note de thèses : Mémoire de master en informatique Machine learning for link prediction in complex networks [texte manuscrit] / Messaoud Babaghayou, Auteur ; Abdallah Lakhdari, Directeur de thèse . - Laghouat : Université Amar Telidji - Département d'informatique, 2016 . - 79 p. ; 30 cm. + 1 disque optique numérique. Option : Networks, systems and distributed applications ( Réseaux,systèmes et applications réparties) Langues : Anglais (eng) Mots-clés : Machine Learning  Link Prediction  Complex Networks  Supervised Leaning  Unsupervised Learning  Classification  Node-based Metrics Résumé : Nowdays, networks are omnipresent. The study and understanding of these networks become a greater need. The purpose of this work, is to investigate link prediction task in complex networks using Machine learning techniques. In fact, we propose two approaches to perform link prediction: supervised and unsupervised one. In both techniques a link or a pair of nodes is characterized by several features based on network topology-based metrics. In addition, we investigate many combined features. Concerning the supervised approach, we investigate the KNN and decision tree methods to build the link prediction models. While in the unsupervised approach, we rely on ranking strategy. An experimental study is performed on real networks. The results show that the supervised approach using gathered features reaches good performances with 84% f-measure. note de thèses : Mémoire de master en informatique

Réservation

Réserver ce document

Exemplaires (1)

Code-barres	Cote	Support	Localisation	Section	Disponibilité
MF 01-12	MF 01-12	Thése	BIBLIOTHEQUE DE FACULTE DES SCIENCES	théses (sci)	Disponible

RAND-WHISPER : a randomized transmission power scheme for enhanced location privacy in the internet of vehicles / Taieb Derrah

Public ISBD Titre : RAND-WHISPER : a randomized transmission power scheme for enhanced location privacy in the internet of vehicles Type de document : document multimédia Auteurs : Taieb Derrah, Auteur ; Messaoud Babaghayou, Directeur de thèse Editeur : Laghouat : Université Amar Telidji - Département d'informatique Année de publication : 2026 Importance : 75 p. Accompagnement : 1 disque optique numérique (CD-ROM) Note générale : Option : Networks, systems and distributed applications Langues : Anglais (eng) Mots-clés : Internet of vehicles  Location privacy  Transmission power control  Randomization  RAND-WHISPER  VANET  OMNeT++, Veins  SUMO  PREXT Résumé : With the rapid growth of the Internet of Vehicles (IoV), vehicles continuously broadcast beacon messages containing sensitive information such as location, speed, and trajectory, exposing them to serious privacy threats including tracking and correlation attacks. Although existing mechanisms such as pseudonym changes, silent periods, and mix zones provide partial protection, they remain vulnerable to pattern-based tracking due to deterministic transmission behavior. This thesis proposes RAND-WHISPER, a randomized transmission power scheme designed to enhance location privacy in IoV environments. Built upon the WHISPER (WSP) scheme, RAND-WHISPER introduces controlled randomness into the transmission power selection by applying a ±ε variation around speed-based power levels, reducing communication pattern predictability and making it significantly harder for adversaries to track vehicles over time. The scheme is implemented and evaluated using OMNeT++, Veins 4.4, and SUMO within the PREXT framework, using a real road network of Tlemcen city, under vehicle densities ranging from 50 to 250. Results demonstrate that RAND-WHISPER achieves improved location privacy compared to WSP and other baselines, while maintaining acceptable communication performance. note de thèses : Mémoire de master en informatique RAND-WHISPER : a randomized transmission power scheme for enhanced location privacy in the internet of vehicles [document multimédia] / Taieb Derrah, Auteur ; Messaoud Babaghayou, Directeur de thèse . - Laghouat : Université Amar Telidji - Département d'informatique, 2026 . - 75 p. + 1 disque optique numérique (CD-ROM). Option : Networks, systems and distributed applications Langues : Anglais (eng) Mots-clés : Internet of vehicles  Location privacy  Transmission power control  Randomization  RAND-WHISPER  VANET  OMNeT++, Veins  SUMO  PREXT Résumé : With the rapid growth of the Internet of Vehicles (IoV), vehicles continuously broadcast beacon messages containing sensitive information such as location, speed, and trajectory, exposing them to serious privacy threats including tracking and correlation attacks. Although existing mechanisms such as pseudonym changes, silent periods, and mix zones provide partial protection, they remain vulnerable to pattern-based tracking due to deterministic transmission behavior. This thesis proposes RAND-WHISPER, a randomized transmission power scheme designed to enhance location privacy in IoV environments. Built upon the WHISPER (WSP) scheme, RAND-WHISPER introduces controlled randomness into the transmission power selection by applying a ±ε variation around speed-based power levels, reducing communication pattern predictability and making it significantly harder for adversaries to track vehicles over time. The scheme is implemented and evaluated using OMNeT++, Veins 4.4, and SUMO within the PREXT framework, using a real road network of Tlemcen city, under vehicle densities ranging from 50 to 250. Results demonstrate that RAND-WHISPER achieves improved location privacy compared to WSP and other baselines, while maintaining acceptable communication performance. note de thèses : Mémoire de master en informatique

---

1 (1 - 5 / 5) Par page : 25 50 100 200