Farid Nait-Abdesselam
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Farid Nait-Abdesselam has a total of 124 co-authors. Here are the top co-authors by number of shared publications:
Ashfaq Khokhar
3 shared publications- Chafiq Titouna
Université Paris Cité, Paris, France
3 shared publications - Anwar Ahmed Khan
Trinity College Dublin
2 shared publications Chafiq Titouna
University of Paris,Paris,France
2 shared publications- Indrakshi Dey
Department of Computing and Mathematics, South East Technological University, Waterford, Ireland
2 shared publications - Khouloud Safi Eljil
LIPADE Laboratory, Paris Cité University, Paris, France
2 shared publications
Latest Publications
- Open Access
- 8 October 2026
- Published by Springer Nature in Cybersecurity
- Vol. 9 (1) , 227
Evasion attacks on generative safeguards: target-concept reappearance under structural control in grey-box settings
Concept erasure has emerged as a practical safeguard for suppressing unsafe or unwanted concepts in text-to-image diffusion models. However, existing methods are primarily designed for text-triggered generation and are usually evaluated under text-only prompting. In this paper, we identify a practical robustness gap by showing that concepts suppressed under text-only evaluation can reappear after a concept-erased backbone is composed with a compatible structural controller and supplied with target-consistent, information-bearing structural conditions. We propose Structure-Guided Concept Reappearance (SGCR), an inference-time attack that combines a benign prompt with a structural condition, such as a Canny edge map, depth map, or segmentation map, to induce target-related outputs without retraining, gradient-based optimization, or modification of the defended backbone. Extensive experiments on explicit content, copyrighted artistic styles, and generic objects demonstrate that SGCR remains effective across multiple representative erasure defenses and structural modalities. These findings show that suppression established under text-only evaluation may not persist in controller-augmented diffusion pipelines. The demonstrated threat applies to local modular systems and services exposing compatible structural-control interfaces, rather than closed text-only APIs.
- 8 October 2026
- Published by Springer Nature in Cybersecurity
- Vol. 9 (1) , 227
Open AccessEvasion attacks on generative safeguards: target-concept reappearance under structural control in grey-box settings
Concept erasure has emerged as a practical safeguard for suppressing unsafe or unwanted concepts in text-to-image diffusion models. However, existing methods are primarily designed for text-triggered generation and are usually evaluated under text-only prompting. In this paper, we identify a practical robustness gap by showing that concepts suppressed under text-only evaluation can reappear after a concept-erased backbone is composed with a compatible structural controller and supplied with target-consistent, information-bearing structural conditions. We propose Structure-Guided Concept Reappearance (SGCR), an inference-time attack that combines a benign prompt with a structural condition, such as a Canny edge map, depth map, or segmentation map, to induce target-related outputs without retraining, gradient-based optimization, or modification of the defended backbone. Extensive experiments on explicit content, copyrighted artistic styles, and generic objects demonstrate that SGCR remains effective across multiple representative erasure defenses and structural modalities. These findings show that suppression established under text-only evaluation may not persist in controller-augmented diffusion pipelines. The demonstrated threat applies to local modular systems and services exposing compatible structural-control interfaces, rather than closed text-only APIs.
- 5 March 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Information Forensics and Security
- Vol. 21 (15566013) , 3575-3587
Trust Under Siege: Label Spoofing Attacks Against Machine Learning for Android Malware Detection
Machine Learning (ML) malware detectors rely heavily on crowd-sourced AntiVirus (AV) labels, with platforms like VirusTotal serving as trusted sources of malware annotations. But what if attackers could manipulate these labels to classify benign software as malicious? We introduce label spoofing attacks, a new threat that contaminates crowd-sourced datasets by embedding minimal and undetectable malicious patterns into benign samples. These patterns coerce AV engines into misclassifying legitimate files as harmful, enabling poisoning attacks against ML-based malware classifiers trained on those data. We demonstrate this scenario by developing AndroVenom, a methodology for polluting realistic data sources and launching subsequent poisoning attacks against ML malware detectors. Experiments show that not only are state-of-the-art feature extractors unable to filter such injections, but various ML models experience Denial-of-Service (DoS) with as little as 1% poisoned samples. Additionally, attackers can flip decisions for specific unaltered benign samples by modifying only 0.015% of the training data, threatening their reputation and market share, while evading anomaly detectors operating on the training data. We conclude by raising concerns about the trustworthiness of ML training processes based on AV annotations and argue that further investigation is needed to develop more reliable labeling strategies.
- 5 March 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Information Forensics and Security
- Vol. 21 (15566013) , 3575-3587
Trust Under Siege: Label Spoofing Attacks Against Machine Learning for Android Malware Detection
Machine Learning (ML) malware detectors rely heavily on crowd-sourced AntiVirus (AV) labels, with platforms like VirusTotal serving as trusted sources of malware annotations. But what if attackers could manipulate these labels to classify benign software as malicious? We introduce label spoofing attacks, a new threat that contaminates crowd-sourced datasets by embedding minimal and undetectable malicious patterns into benign samples. These patterns coerce AV engines into misclassifying legitimate files as harmful, enabling poisoning attacks against ML-based malware classifiers trained on those data. We demonstrate this scenario by developing AndroVenom, a methodology for polluting realistic data sources and launching subsequent poisoning attacks against ML malware detectors. Experiments show that not only are state-of-the-art feature extractors unable to filter such injections, but various ML models experience Denial-of-Service (DoS) with as little as 1% poisoned samples. Additionally, attackers can flip decisions for specific unaltered benign samples by modifying only 0.015% of the training data, threatening their reputation and market share, while evading anomaly detectors operating on the training data. We conclude by raising concerns about the trustworthiness of ML training processes based on AV annotations and argue that further investigation is needed to develop more reliable labeling strategies.
- Open Access
Next-Generation MAC Technique for Priority Handling in Industrial Cyber-Physical Systems
Next Generation Media Access Control (NGMA) techniques have been designed to support diverse applications with heterogeneous priorities. In industrial cyber-physical systems (CPS), the number of connected devices and systems is expected to grow significantly, demanding dependable and prompt network services. In this work, we present a novel scheme, Dynamic Fragmentation-MAC (DyFrag-MAC) that offers dynamic, differentiated channel access to the traffic of various priorities. DyFrag-MAC works on fragmenting the data of normal priority in order to support early delivery of urgent priority data. In prior work, urgent priority data either had to wait for the complete transmission of lower-priority packets or relied on multi-channel protocols to gain access. We compared the proposed fragmentation scheme with FROG-MAC and industrial Deterministic and Synchronous Multi-channel Extension (i-DSME). FROG-MAC fragmented the lower priority packets, but did not adjust the fragment size dynamically, whereas i-DSME utilized multiple channels and adaptive contention mechanisms; both protocols lack the ability to preempt ongoing lower-priority transmissions. Hence, the performance evaluation in terms of average delay and throughput reveals better performance of DyFRAG-MAC for the heterogeneous traffic.
Open AccessNext-Generation MAC Technique for Priority Handling in Industrial Cyber-Physical Systems
Next Generation Media Access Control (NGMA) techniques have been designed to support diverse applications with heterogeneous priorities. In industrial cyber-physical systems (CPS), the number of connected devices and systems is expected to grow significantly, demanding dependable and prompt network services. In this work, we present a novel scheme, Dynamic Fragmentation-MAC (DyFrag-MAC) that offers dynamic, differentiated channel access to the traffic of various priorities. DyFrag-MAC works on fragmenting the data of normal priority in order to support early delivery of urgent priority data. In prior work, urgent priority data either had to wait for the complete transmission of lower-priority packets or relied on multi-channel protocols to gain access. We compared the proposed fragmentation scheme with FROG-MAC and industrial Deterministic and Synchronous Multi-channel Extension (i-DSME). FROG-MAC fragmented the lower priority packets, but did not adjust the fragment size dynamically, whereas i-DSME utilized multiple channels and adaptive contention mechanisms; both protocols lack the ability to preempt ongoing lower-priority transmissions. Hence, the performance evaluation in terms of average delay and throughput reveals better performance of DyFRAG-MAC for the heterogeneous traffic.
- 25 September 2025
- Published by Elsevier in Ad Hoc Networks
- Vol. 180
OCUS: A game-theoretic approach to optimal UAV coalitions in UAV-as-a-Service platforms
- 25 September 2025
- Published by Elsevier in Ad Hoc Networks
- Vol. 180
OCUS: A game-theoretic approach to optimal UAV coalitions in UAV-as-a-Service platforms
- 19 August 2025
- Published by Springer Nature in Cluster Computing
- Vol. 28 (8) , 1-17
Efficient shard load balancing for scalable blockchain systems
- 19 August 2025
- Published by Springer Nature in Cluster Computing
- Vol. 28 (8) , 1-17
Efficient shard load balancing for scalable blockchain systems
- 15 July 2025
- Published by Springer Nature
- p. 390-401
Enhancing Privacy Guarantees for Federated Learning with Local Differential Privacy and Secure Multi-party Computation
- 15 July 2025
- Published by Springer Nature
- p. 390-401
Enhancing Privacy Guarantees for Federated Learning with Local Differential Privacy and Secure Multi-party Computation
- Open Access
- 24 April 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in TechRxiv
Spotting Road's Dangers: Graph Neural Networks for Abnormal Driving Detection
Abnormal driving behaviors, such as aggressive and impaired driving, pose significant risks to road safety for both drivers and pedestrians. Vehicular networks, with their ability to enable real-time communication between vehicles and infrastructure, offer promising avenues for detecting and mitigating these risks. This work leverages Cooperative Awareness Messages (CAMs) and synthetic datasets to model intervehicular relationships through network topologies, enabling effective behavior monitoring. We utilize supervised Graph Neural Networks (GNNs), including Graph Convolutional Network (GCN), Graph Attention Network (GAT), Graph Sample and Aggregate (GraphSAGE), Temporal Graph Convolutional Network (T-GCN), Graph Convolutional Long Short-Term Memory (GConvLSTM), and Graph Convolutional Gated Recurrent Unit (GConvGRU), to detect abnormal driving patterns by capturing their topological features. To enhance detection accuracy and robustness, we introduce two key methods: (1) Edge weight learning, which dynamically assigns edge weights based on node features, improving the model's ability to focus on critical connections, and (2) a supervised triplet loss mechanism, which clusters nodes of similar classes while increasing separation between different classes for better classification. We further address class imbalance by applying class weighting and the Synthetic Minority Oversampling Technique (SMOTE) to ensure balanced model training. Experimental results demonstrate that GNNs, when equipped with these enhancements, can effectively identify dangerous driving behaviors with high accuracy, underscoring their potential for real-time safety monitoring in vehicular networks.
- 24 April 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in TechRxiv
Open AccessSpotting Road's Dangers: Graph Neural Networks for Abnormal Driving Detection
Abnormal driving behaviors, such as aggressive and impaired driving, pose significant risks to road safety for both drivers and pedestrians. Vehicular networks, with their ability to enable real-time communication between vehicles and infrastructure, offer promising avenues for detecting and mitigating these risks. This work leverages Cooperative Awareness Messages (CAMs) and synthetic datasets to model intervehicular relationships through network topologies, enabling effective behavior monitoring. We utilize supervised Graph Neural Networks (GNNs), including Graph Convolutional Network (GCN), Graph Attention Network (GAT), Graph Sample and Aggregate (GraphSAGE), Temporal Graph Convolutional Network (T-GCN), Graph Convolutional Long Short-Term Memory (GConvLSTM), and Graph Convolutional Gated Recurrent Unit (GConvGRU), to detect abnormal driving patterns by capturing their topological features. To enhance detection accuracy and robustness, we introduce two key methods: (1) Edge weight learning, which dynamically assigns edge weights based on node features, improving the model's ability to focus on critical connections, and (2) a supervised triplet loss mechanism, which clusters nodes of similar classes while increasing separation between different classes for better classification. We further address class imbalance by applying class weighting and the Synthetic Minority Oversampling Technique (SMOTE) to ensure balanced model training. Experimental results demonstrate that GNNs, when equipped with these enhancements, can effectively identify dangerous driving behaviors with high accuracy, underscoring their potential for real-time safety monitoring in vehicular networks.
- Open Access
- 21 April 2025
- Published by Springer Nature in International Journal of Networked and Distributed Computing
- Vol. 13 (1) , 1-17
Use of Asymmetric Byzantine Quorums to Overcome Trust Issues in Blockchain Oracles
- 21 April 2025
- Published by Springer Nature in International Journal of Networked and Distributed Computing
- Vol. 13 (1) , 1-17
Open AccessUse of Asymmetric Byzantine Quorums to Overcome Trust Issues in Blockchain Oracles
- 24 February 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 12 (12) , 20870-20884
Enhancing UAV Network Security: A Human-in-the-Loop and GAN-Based Approach to Intrusion Detection
Unmanned Aerial Vehicles (UAVs) are becoming essential in various sectors such as commercial delivery, agricultural monitoring, and disaster response. Despite their benefits, the rapid adoption of UAVs poses substantial security challenges, especially in drone network intrusion detection. Traditional intrusion detection datasets often suffer from limitations like small sample sizes and uneven distribution, undermining the effectiveness of Intrusion Detection Systems (IDS). Moreover, conventional machine learning (ML) approaches generally require extensive, well-labeled datasets that are expensive and labor intensive to produce. To overcome these challenges, we introduce a Generative Adversarial Network (GAN) model designed to enhance and balance the limited datasets available for drone networks. This model significantly improves data quality and quantity, thus optimizing the training process for intrusion detection models. Furthermore, we propose a Human-in-the-Loop (HITL) ML framework that integrates human expertise to guide the learning process and mitigate the costs of labeling. Our comprehensive evaluation demonstrates that the combined application of the GAN model and the HITL framework significantly outperforms traditional baseline models. This approach not only achieves an intrusion detection accuracy of up to 99% across various experimental datasets but also dramatically reduces the requirement for large amounts of labeled data by up to 98%, providing a cost-effective solution for enhancing UAV network security.
- 24 February 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 12 (12) , 20870-20884
Enhancing UAV Network Security: A Human-in-the-Loop and GAN-Based Approach to Intrusion Detection
Unmanned Aerial Vehicles (UAVs) are becoming essential in various sectors such as commercial delivery, agricultural monitoring, and disaster response. Despite their benefits, the rapid adoption of UAVs poses substantial security challenges, especially in drone network intrusion detection. Traditional intrusion detection datasets often suffer from limitations like small sample sizes and uneven distribution, undermining the effectiveness of Intrusion Detection Systems (IDS). Moreover, conventional machine learning (ML) approaches generally require extensive, well-labeled datasets that are expensive and labor intensive to produce. To overcome these challenges, we introduce a Generative Adversarial Network (GAN) model designed to enhance and balance the limited datasets available for drone networks. This model significantly improves data quality and quantity, thus optimizing the training process for intrusion detection models. Furthermore, we propose a Human-in-the-Loop (HITL) ML framework that integrates human expertise to guide the learning process and mitigate the costs of labeling. Our comprehensive evaluation demonstrates that the combined application of the GAN model and the HITL framework significantly outperforms traditional baseline models. This approach not only achieves an intrusion detection accuracy of up to 99% across various experimental datasets but also dramatically reduces the requirement for large amounts of labeled data by up to 98%, providing a cost-effective solution for enhancing UAV network security.
- 27 January 2025
- Published by Springer Nature in The Journal of Supercomputing
- Vol. 81 (2) , 1-35
An enhanced mechanism for malicious URL detection using deep learning and DistilBERT-based feature extraction
- 27 January 2025
- Published by Springer Nature in The Journal of Supercomputing
- Vol. 81 (2) , 1-35
An enhanced mechanism for malicious URL detection using deep learning and DistilBERT-based feature extraction