As part of the ENSURE-6G project, Professor An Braeken and Abdessamad Nassihi from Vrije Universiteit Brussel (VUB) undertook a research secondment at Montimage (MON) in Paris, France. Hosted by Edgardo Montes de Oca, Wissam Mallouli, and Ana Cavalli, the secondment provided an excellent environment for knowledge exchange, collaboration, and the exploration of advanced topics in AI-driven network security and hardware acceleration.
Bridging Software and Hardware Approaches to Intrusion Detection
During the visit, Abdessamad presented and discussed his recent master’s thesis work on machine learning (ML)-based Intrusion Detection Systems (IDS) for the Internet of Things (IoT). His research combined multiple public datasets into a unified benchmark and performed an in-depth evaluation of various ML models. The best-performing model was then implemented on an FPGA, using both HLS4ML and hand-written VHDL approaches.
This work introduced a hardware-oriented perspective to Montimage’s advanced monitoring and analysis ecosystem. The company’s solutions, the Montimage AI Platform (MAIP) and MMT Probe, form a robust framework for AI-driven intrusion detection and network traffic analysis. MAIP provides an end-to-end environment for data preprocessing, feature extraction, model training, and explainability visualization, while MMT Probe delivers high-performance packet analysis, capable of handling extensive protocol-level features.
Combining these software capabilities with a hardware-accelerated IDS approach led to insightful discussions on how future security frameworks can achieve lower latency, greater energy efficiency, and enhanced scalability at the edge. These exchanges also touched on the importance of trustworthy AI and explainable security mechanisms, particularly through the use of tools and techniques that make it possible to interpret and justify model decisions. Together, we explored how hardware acceleration and explainable AI can intersect to create resilient, transparent, and power-efficient network protection mechanisms for next-generation communication systems. This synergy between hardware efficiency and AI intelligence became a central theme of the secondment and opened promising directions for continued research.
Advancing Research Through a Comprehensive Survey
A key outcome of the secondment was the collaborative work on a comprehensive survey paper. The study critically examines the limitations of current ML-based IDS, focusing not only on algorithmic performance but also on the methodological and data challenges that hinder progress in the field.
Through the survey, we discussed one of the major obstacles facing current ML-based IDS solutions, namely the difficulty of obtaining realistic datasets that accurately capture modern network behavior. Most available datasets are created in isolated or simulated environments and therefore fail to represent the diversity and complexity of real network traffic. Another limitation is that IDS models trained on one dataset rarely generalize to others, showing strong dependence on dataset-specific features rather than on genuine malicious behavior.
To address these limitations, the paper introduces a comparability first approach by re evaluating representative IDS models within a unified framework that respects each dataset’s native feature space. This method provides a fair assessment of model generalization under diverse traffic conditions and reveals the gap between dataset-specific accuracy and real-world robustness. The work was developed under the guidance of Professor An Braeken and in close collaboration with researchers at Montimage, whose expertise and feedback were essential in shaping the methodology and refining the overall analysis.
The survey also emphasizes emerging directions such as standardized evaluation protocols, cross-dataset benchmarking, and the integration of explainable AI for greater transparency in detection decisions. Targeted for submission to IEEE Communications Surveys & Tutorials, this work contributes directly to ENSURE-6G Task 2.1 and Deliverable D2.1, both of which focus on trustworthy and distributed AI in 6G ecosystems.
Reflections on the Secondment
The secondment at Montimage was both a professional and personal milestone. It offered valuable insight into how AI and hardware acceleration can be jointly applied to secure next generation networks. Working closely with experts in network monitoring and AI-driven security not only expanded our technical perspective but also demonstrated the importance of interdisciplinary collaboration in advancing 6G research.
