Research interests
Current work sits between host-based intrusion detection and quantum machine learning. The focus is on detecting memory-resident attacks, and on testing whether quantum kernel methods improve on classical baselines.
MemoryWatch
In progressA host-based intrusion detection system for memory-based attacks, built as a team project in EPITA's Action Learning programme. It compares a classical anomaly detector, Isolation Forest, against ZZFeatureMap quantum kernels. Both are tested on simulated memory attack datasets and on real-world malware data from CIC-MalMem-2022. The open question is whether quantum feature-space separation gives a measurable advantage in multiclass attack classification.
What the research covers
Quantum-inspired anomaly detection
Applies quantum kernel methods to intrusion detection and compares them with classical machine learning. MemoryWatch is the main test bed for this, running both approaches against the same attack data.
Memory-based threat detection
Covers fileless techniques such as heap spraying, injection through the Linux /proc filesystem and memory-resident malware. These attacks often leave little on disk, which lets them evade signature-based tools. The work uses kernel-level behaviour, such as syscall frequency and /proc access patterns, as detection signals.
Host-based intrusion detection
Builds unsupervised anomaly detection pipelines that do not depend on labelled attack data. This includes feature engineering from kernel traces and dynamic threshold calibration. Results are reported with FPR, recall, F1 and AUROC rather than accuracy alone, since accuracy can hide weak detection on imbalanced data.
Quantum computing for security in the NISQ era
Studies the practical limits of quantum machine learning on current hardware: gate noise, short coherence times and the gap between simulation and real devices. Experiments on real IBM quantum hardware are planned from late 2026.