Slide
Centre Interdisciplinaire
de Recherche et d’Innovation
en Cybersécurité et Société
de Recherche et d’Innovation
en Cybersécurité et Société
1.
Ameyoud, S. Mohamed; Allili, M. Saïd
Multi-modal malware classification with hierarchical consistency and saliency-constrained adversarial training Journal Article
In: Journal of Information Security and Applications, vol. 99, 2026, ISSN: 22142134 (ISSN).
Abstract | Links | BibTeX | Tags: Adversarial training, Capability of detection, Classification (of information), Convolution, convolutional neural network, Convolutional neural networks, Detection system, Hierarchical consistency, Hierarchical systems, Malware, Malware classification, Malware classifications, Malware families, Malwares, Multi-modal, Multi-modal learning, Semantics, Vision transformer, Vision transformers
@article{mohamed_ameyoud_multi-modal_2026,
title = {Multi-modal malware classification with hierarchical consistency and saliency-constrained adversarial training},
author = {S. Mohamed Ameyoud and M. Saïd Allili},
url = {https://www.scopus.com/inward/record.uri?eid=2-s2.0-105031186108&doi=10.1016%2Fj.jisa.2026.104429&partnerID=40&md5=2425da4ab40f9043ba4e67d223a1bdd9},
doi = {10.1016/j.jisa.2026.104429},
issn = {22142134 (ISSN)},
year = {2026},
date = {2026-01-01},
journal = {Journal of Information Security and Applications},
volume = {99},
abstract = {The increasing complexity of malware, including polymorphic, obfuscated, and adversarial variants, continues to outpace the capabilities of detection systems. Here, we introduce a robust multi-modal hierarchical framework that jointly leverages visual and code-level semantics to enhance malware family and type classification. Our architecture fuses convolutional and transformer-based encoders to extract complementary representations from raw malware binaries and decompiled control-flow functions, enabling a rich, cross-modal understanding of malicious behavior. The classification pipeline follows a two-stage hierarchical protocol, where the predicted malware type informs the family-level classification. This enforces ontological consistency between type and family prediction levels. To further bolster robustness against adversarial and obfuscated malware, we integrate a novel adversarial training strategy that generates plausible perturbations guided by attention distributions. Evaluation on multiple large-scale benchmarks including BODMAS, Malimg, Microsoft BIG 2015, and a curated set of from MalwareBazaar, demonstrate that our framework consistently outperforms state-of-the-art baselines, including ResNet, Swin Transformer, and MalBERTv2, across both malware type and family prediction tasks. Notably, our model exhibits outstanding generalization to unpacked, obfuscated, and previously unseen samples, with minimal performance degradation. It achieves accuracy gains of +3-6% over leading methods and exhibits superior resilience under adversarial threat models. These results highlight the effectiveness of hierarchical conditioning, adversarial robustness, and multi-modal fusion in tackling the evolving landscape of malware. The proposed framework thus offers a scalable and generalizable approach for next-generation malware classification in real-world cybersecurity environments. © 2026 Elsevier Ltd.},
keywords = {Adversarial training, Capability of detection, Classification (of information), Convolution, convolutional neural network, Convolutional neural networks, Detection system, Hierarchical consistency, Hierarchical systems, Malware, Malware classification, Malware classifications, Malware families, Malwares, Multi-modal, Multi-modal learning, Semantics, Vision transformer, Vision transformers},
pubstate = {published},
tppubtype = {article}
}
The increasing complexity of malware, including polymorphic, obfuscated, and adversarial variants, continues to outpace the capabilities of detection systems. Here, we introduce a robust multi-modal hierarchical framework that jointly leverages visual and code-level semantics to enhance malware family and type classification. Our architecture fuses convolutional and transformer-based encoders to extract complementary representations from raw malware binaries and decompiled control-flow functions, enabling a rich, cross-modal understanding of malicious behavior. The classification pipeline follows a two-stage hierarchical protocol, where the predicted malware type informs the family-level classification. This enforces ontological consistency between type and family prediction levels. To further bolster robustness against adversarial and obfuscated malware, we integrate a novel adversarial training strategy that generates plausible perturbations guided by attention distributions. Evaluation on multiple large-scale benchmarks including BODMAS, Malimg, Microsoft BIG 2015, and a curated set of from MalwareBazaar, demonstrate that our framework consistently outperforms state-of-the-art baselines, including ResNet, Swin Transformer, and MalBERTv2, across both malware type and family prediction tasks. Notably, our model exhibits outstanding generalization to unpacked, obfuscated, and previously unseen samples, with minimal performance degradation. It achieves accuracy gains of +3-6% over leading methods and exhibits superior resilience under adversarial threat models. These results highlight the effectiveness of hierarchical conditioning, adversarial robustness, and multi-modal fusion in tackling the evolving landscape of malware. The proposed framework thus offers a scalable and generalizable approach for next-generation malware classification in real-world cybersecurity environments. © 2026 Elsevier Ltd.
2.
Ameyoud, S. Mohamed; Allili, M. Saïd
Multi-modal malware classification with hierarchical consistency and saliency-constrained adversarial training Journal Article
In: Journal of Information Security and Applications, vol. 99, 2026, ISSN: 22142134 (ISSN).
Abstract | Links | BibTeX | Tags: Adversarial training, Capability of detection, Classification (of information), Convolution, convolutional neural network, Convolutional neural networks, Detection system, Hierarchical consistency, Hierarchical systems, Malware, Malware classification, Malware classifications, Malware families, Malwares, Multi-modal, Multi-modal learning, Semantics, Vision transformer, Vision transformers
@article{mohamedameyoudMultimodalMalwareClassification2026,
title = {Multi-modal malware classification with hierarchical consistency and saliency-constrained adversarial training},
author = {S. Mohamed Ameyoud and M. Saïd Allili},
url = {https://www.scopus.com/pages/publications/105031186108?origin=resultslist},
doi = {10.1016/j.jisa.2026.104429},
issn = {22142134 (ISSN)},
year = {2026},
date = {2026-01-01},
journal = {Journal of Information Security and Applications},
volume = {99},
publisher = {Elsevier Ltd},
abstract = {The increasing complexity of malware, including polymorphic, obfuscated, and adversarial variants, continues to outpace the capabilities of detection systems. Here, we introduce a robust multi-modal hierarchical framework that jointly leverages visual and code-level semantics to enhance malware family and type classification. Our architecture fuses convolutional and transformer-based encoders to extract complementary representations from raw malware binaries and decompiled control-flow functions, enabling a rich, cross-modal understanding of malicious behavior. The classification pipeline follows a two-stage hierarchical protocol, where the predicted malware type informs the family-level classification. This enforces ontological consistency between type and family prediction levels. To further bolster robustness against adversarial and obfuscated malware, we integrate a novel adversarial training strategy that generates plausible perturbations guided by attention distributions. Evaluation on multiple large-scale benchmarks including BODMAS, Malimg, Microsoft BIG 2015, and a curated set of from MalwareBazaar, demonstrate that our framework consistently outperforms state-of-the-art baselines, including ResNet, Swin Transformer, and MalBERTv2, across both malware type and family prediction tasks. Notably, our model exhibits outstanding generalization to unpacked, obfuscated, and previously unseen samples, with minimal performance degradation. It achieves accuracy gains of +3-6% over leading methods and exhibits superior resilience under adversarial threat models. These results highlight the effectiveness of hierarchical conditioning, adversarial robustness, and multi-modal fusion in tackling the evolving landscape of malware. The proposed framework thus offers a scalable and generalizable approach for next-generation malware classification in real-world cybersecurity environments. © 2026 Elsevier Ltd.},
keywords = {Adversarial training, Capability of detection, Classification (of information), Convolution, convolutional neural network, Convolutional neural networks, Detection system, Hierarchical consistency, Hierarchical systems, Malware, Malware classification, Malware classifications, Malware families, Malwares, Multi-modal, Multi-modal learning, Semantics, Vision transformer, Vision transformers},
pubstate = {published},
tppubtype = {article}
}
The increasing complexity of malware, including polymorphic, obfuscated, and adversarial variants, continues to outpace the capabilities of detection systems. Here, we introduce a robust multi-modal hierarchical framework that jointly leverages visual and code-level semantics to enhance malware family and type classification. Our architecture fuses convolutional and transformer-based encoders to extract complementary representations from raw malware binaries and decompiled control-flow functions, enabling a rich, cross-modal understanding of malicious behavior. The classification pipeline follows a two-stage hierarchical protocol, where the predicted malware type informs the family-level classification. This enforces ontological consistency between type and family prediction levels. To further bolster robustness against adversarial and obfuscated malware, we integrate a novel adversarial training strategy that generates plausible perturbations guided by attention distributions. Evaluation on multiple large-scale benchmarks including BODMAS, Malimg, Microsoft BIG 2015, and a curated set of from MalwareBazaar, demonstrate that our framework consistently outperforms state-of-the-art baselines, including ResNet, Swin Transformer, and MalBERTv2, across both malware type and family prediction tasks. Notably, our model exhibits outstanding generalization to unpacked, obfuscated, and previously unseen samples, with minimal performance degradation. It achieves accuracy gains of +3-6% over leading methods and exhibits superior resilience under adversarial threat models. These results highlight the effectiveness of hierarchical conditioning, adversarial robustness, and multi-modal fusion in tackling the evolving landscape of malware. The proposed framework thus offers a scalable and generalizable approach for next-generation malware classification in real-world cybersecurity environments. © 2026 Elsevier Ltd.
3.
Malasi, J. -J. M.; Moudoud, H.; Missaoui, R.
A Lightweight Multimodal LLM-Based Intrusion Detection System for Open RAN Proceedings Article
In: IEEE Wirel. Commun. Netw. Conf. Workshops, WCNCW, Institute of Electrical and Electronics Engineers Inc., 2026, ISBN: 979-833157731-5 (ISBN), (Journal Abbreviation: IEEE Wirel. Commun. Netw. Conf. Workshops, WCNCW).
Abstract | Links | BibTeX | Tags: Benchmarking, Computational linguistics, Computer crime, Cyber security, Cybersecurity, Embedded systems, Embeddings, Intelligent controllers, Internet protocols, Intrusion Detection, intrusion detection system, Intrusion Detection System (IDS), Intrusion Detection Systems, Language model, Large language model, Large Language Models (LLMs), Mobile telecommunication systems, Multi-modal, Multi-modal learning, Multimodal Learning, Network security, Open RAN, Pipelines, RAN intelligent controller, RAN intelligent controller (RIC), Semantics
@inproceedings{malasiLightweightMultimodalLLMBased2026,
title = {A Lightweight Multimodal LLM-Based Intrusion Detection System for Open RAN},
author = {J. -J. M. Malasi and H. Moudoud and R. Missaoui},
url = {https://www.scopus.com/pages/publications/105043415034?origin=resultslist},
doi = {10.1109/WCNCW67598.2026.11555393},
isbn = {979-833157731-5 (ISBN)},
year = {2026},
date = {2026-01-01},
booktitle = {IEEE Wirel. Commun. Netw. Conf. Workshops, WCNCW},
publisher = {Institute of Electrical and Electronics Engineers Inc.},
abstract = {Open Radio Access Network architectures introduce unprecedented openness and programmability through RAN Intelligent Controllers, expanding the attack surface beyond traditional volumetric threats. Most existing intrusion detection systems for fifth-generation mobile networks and ORAN rely primarily on numerical Key Performance Indicators (KPIs), often overlooking the security-relevant semantics embedded in logs and control messages. This paper proposes LLM4IDS, a semantic-aware and multimodal intrusion detection system that fuses lightweight Large Language Model (LLM) embeddings of textified events with traditional tabular KPIs in a compact Transformer-based classifier. Evaluated on the realworld OpenIreland O-RAN dataset and two classical IP network benchmarks (UNSW-NB15 and CICIDS2017), LLM4IDS consistently matches or surpasses strong tabular baselines while maintaining high efficiency. On OpenIreland, it reaches nearperfect detection (F1-score $textbackslashapprox 1.0$) and yields large gains for application-layer and volumetric attacks compared to KPI-only models. Across datasets, the classifier maintains an approximately 4 MB footprint and sub-millisecond CPU inference latency; the frozen sentence-LLM encoder can be shared across tasks, and its embeddings can be cached or precomputed when required by the deployment pipeline. These results indicate that integrating semantic context through multimodal fusion can significantly enhance intrusion detection in O-RAN while remaining competitive on standard IP-based IDS tasks. To the best of our knowledge, this work is among the first to study a complete multimodal LLM-based IDS pipeline for O-RAN data with cross-domain evaluation on both O-RAN and classical IP benchmarks. © 2026 IEEE.},
note = {Journal Abbreviation: IEEE Wirel. Commun. Netw. Conf. Workshops, WCNCW},
keywords = {Benchmarking, Computational linguistics, Computer crime, Cyber security, Cybersecurity, Embedded systems, Embeddings, Intelligent controllers, Internet protocols, Intrusion Detection, intrusion detection system, Intrusion Detection System (IDS), Intrusion Detection Systems, Language model, Large language model, Large Language Models (LLMs), Mobile telecommunication systems, Multi-modal, Multi-modal learning, Multimodal Learning, Network security, Open RAN, Pipelines, RAN intelligent controller, RAN intelligent controller (RIC), Semantics},
pubstate = {published},
tppubtype = {inproceedings}
}
Open Radio Access Network architectures introduce unprecedented openness and programmability through RAN Intelligent Controllers, expanding the attack surface beyond traditional volumetric threats. Most existing intrusion detection systems for fifth-generation mobile networks and ORAN rely primarily on numerical Key Performance Indicators (KPIs), often overlooking the security-relevant semantics embedded in logs and control messages. This paper proposes LLM4IDS, a semantic-aware and multimodal intrusion detection system that fuses lightweight Large Language Model (LLM) embeddings of textified events with traditional tabular KPIs in a compact Transformer-based classifier. Evaluated on the realworld OpenIreland O-RAN dataset and two classical IP network benchmarks (UNSW-NB15 and CICIDS2017), LLM4IDS consistently matches or surpasses strong tabular baselines while maintaining high efficiency. On OpenIreland, it reaches nearperfect detection (F1-score $textbackslashapprox 1.0$) and yields large gains for application-layer and volumetric attacks compared to KPI-only models. Across datasets, the classifier maintains an approximately 4 MB footprint and sub-millisecond CPU inference latency; the frozen sentence-LLM encoder can be shared across tasks, and its embeddings can be cached or precomputed when required by the deployment pipeline. These results indicate that integrating semantic context through multimodal fusion can significantly enhance intrusion detection in O-RAN while remaining competitive on standard IP-based IDS tasks. To the best of our knowledge, this work is among the first to study a complete multimodal LLM-based IDS pipeline for O-RAN data with cross-domain evaluation on both O-RAN and classical IP benchmarks. © 2026 IEEE.
4.
Ameyoud, S. Mohamed; Allili, M. Saïd
Multi-modal malware classification with hierarchical consistency and saliency-constrained adversarial training Journal Article
In: Journal of Information Security and Applications, vol. 99, 2026, ISSN: 22142134 (ISSN).
Abstract | Links | BibTeX | Tags: Adversarial training, Capability of detection, Classification (of information), Convolution, convolutional neural network, Convolutional neural networks, Detection system, Hierarchical consistency, Hierarchical systems, Malware, Malware classification, Malware classifications, Malware families, Malwares, Multi-modal, Multi-modal learning, Semantics, Vision transformer, Vision transformers
@article{mohamedameyoudMultimodalMalwareClassification2026a,
title = {Multi-modal malware classification with hierarchical consistency and saliency-constrained adversarial training},
author = {S. Mohamed Ameyoud and M. Saïd Allili},
url = {https://www.scopus.com/inward/record.uri?eid=2-s2.0-105031186108&doi=10.1016%2Fj.jisa.2026.104429&partnerID=40&md5=2425da4ab40f9043ba4e67d223a1bdd9},
doi = {10.1016/j.jisa.2026.104429},
issn = {22142134 (ISSN)},
year = {2026},
date = {2026-01-01},
journal = {Journal of Information Security and Applications},
volume = {99},
abstract = {The increasing complexity of malware, including polymorphic, obfuscated, and adversarial variants, continues to outpace the capabilities of detection systems. Here, we introduce a robust multi-modal hierarchical framework that jointly leverages visual and code-level semantics to enhance malware family and type classification. Our architecture fuses convolutional and transformer-based encoders to extract complementary representations from raw malware binaries and decompiled control-flow functions, enabling a rich, cross-modal understanding of malicious behavior. The classification pipeline follows a two-stage hierarchical protocol, where the predicted malware type informs the family-level classification. This enforces ontological consistency between type and family prediction levels. To further bolster robustness against adversarial and obfuscated malware, we integrate a novel adversarial training strategy that generates plausible perturbations guided by attention distributions. Evaluation on multiple large-scale benchmarks including BODMAS, Malimg, Microsoft BIG 2015, and a curated set of from MalwareBazaar, demonstrate that our framework consistently outperforms state-of-the-art baselines, including ResNet, Swin Transformer, and MalBERTv2, across both malware type and family prediction tasks. Notably, our model exhibits outstanding generalization to unpacked, obfuscated, and previously unseen samples, with minimal performance degradation. It achieves accuracy gains of +3-6% over leading methods and exhibits superior resilience under adversarial threat models. These results highlight the effectiveness of hierarchical conditioning, adversarial robustness, and multi-modal fusion in tackling the evolving landscape of malware. The proposed framework thus offers a scalable and generalizable approach for next-generation malware classification in real-world cybersecurity environments. © 2026 Elsevier Ltd.},
keywords = {Adversarial training, Capability of detection, Classification (of information), Convolution, convolutional neural network, Convolutional neural networks, Detection system, Hierarchical consistency, Hierarchical systems, Malware, Malware classification, Malware classifications, Malware families, Malwares, Multi-modal, Multi-modal learning, Semantics, Vision transformer, Vision transformers},
pubstate = {published},
tppubtype = {article}
}
The increasing complexity of malware, including polymorphic, obfuscated, and adversarial variants, continues to outpace the capabilities of detection systems. Here, we introduce a robust multi-modal hierarchical framework that jointly leverages visual and code-level semantics to enhance malware family and type classification. Our architecture fuses convolutional and transformer-based encoders to extract complementary representations from raw malware binaries and decompiled control-flow functions, enabling a rich, cross-modal understanding of malicious behavior. The classification pipeline follows a two-stage hierarchical protocol, where the predicted malware type informs the family-level classification. This enforces ontological consistency between type and family prediction levels. To further bolster robustness against adversarial and obfuscated malware, we integrate a novel adversarial training strategy that generates plausible perturbations guided by attention distributions. Evaluation on multiple large-scale benchmarks including BODMAS, Malimg, Microsoft BIG 2015, and a curated set of from MalwareBazaar, demonstrate that our framework consistently outperforms state-of-the-art baselines, including ResNet, Swin Transformer, and MalBERTv2, across both malware type and family prediction tasks. Notably, our model exhibits outstanding generalization to unpacked, obfuscated, and previously unseen samples, with minimal performance degradation. It achieves accuracy gains of +3-6% over leading methods and exhibits superior resilience under adversarial threat models. These results highlight the effectiveness of hierarchical conditioning, adversarial robustness, and multi-modal fusion in tackling the evolving landscape of malware. The proposed framework thus offers a scalable and generalizable approach for next-generation malware classification in real-world cybersecurity environments. © 2026 Elsevier Ltd.