PhD Defense, Abdelhak Hidouri – 15 April, 2025

Abdelhak, Ph.D. student in collaboration with National School for Computer Science (ENSI) – Tunisia, will defend his Ph.D. thesis on April15, 2025, at 1.30 p.m, in the room 37.4.43 at Cnam, 2 rue Conté, 75003.

PhD Title: Intelligent Approach to Improve the Security of Named Data Networks

PhD Title (French): Approche intelligente pour améliorer la sécurité de l’architecture Named Data Networking

Abstract

Named Data Networking (NDN) represents a paradigm shift in internet architecture, offering a data-centric approach that addresses fundamental limitations of the current IP-based infrastructure. By focusing on what users want rather than where data resides, NDN facilitates efficient content distribution, enhanced security, and improved resilience. NDN’s inherent support for caching, built-in security model, graceful handling of intermittent connectivity, and name-based routing offer significant advantages for modern networking, particularly in the context of streaming media, IoT, and Edge Computing.

Despite these benefits, NDN faces critical security challenges, most notably Cache Pollution Attacks (CPAs). CPAs exploit NDN’s caching mechanisms by injecting unpopular content, degrading cache performance and negatively impacting legitimate users. This thesis addresses the urgent need for robust CPA detection and mitigation strategies. It begins with a comprehensive state-of-the-art review of existing techniques, analyzing their strengths and weaknesses, and highlighting key research gaps. A detailed impact analysis of CPAs on NDN performance is also presented, underscoring the necessity for more effective solutions.

To address these challenges, this thesis proposes a series of novel CPA detection and mitigation mechanisms. First, a statistically-based solution, ICAN (Intrusion detection system for CPA Attacks in NDN), is introduced. ICAN leverages key network metrics, including Cache Hit Ratio, Interest Inter-Arrival Time, hop count variation, and prefix variations, for lightweight, and accurate attack detection and mitigation. Second,

recognizing the limitations of ICAN’s reliance on predefined thresholds, a Q-Learning-based solution, Q-ICAN, is developed. Q-ICAN employs reinforcement learning agents within NDN routers to autonomously and adaptively detect and mitigate CPAs, learning from network traffic patterns and dynamically adjusting to changing conditions. This approach proactively mitigates attacks at the interest packet level, a key differentiator from existing solutions.

Finally, to further enhance performance and address the limitations of Q-ICAN in complex, high-dimensional environments, Deep Q-ICAN, a Deep Reinforcement Learning-based solution, is presented. Deep Q-ICAN utilizes Deep Neural Networks to approximate Q-values, enabling efficient handling of large state spaces and continuous action streams. Extensive simulations across various real-world topologies demonstrate the significant performance improvements achieved by Deep Q-ICAN. Specifically, Deep Q-ICAN achieves a 98% Cache Hit Ratio, reduces Average Retrieval Delay to 0.065s, and achieves a CPA detection accuracy of 98.87%, outperforming state-of-the-art solutions and representing a significant advancement in NDN security.

Abdelhak’s publications