Research

Our published work and technical reports on AI, cybersecurity, and decision-making.

Research Topics

Decision Transformers & Offline RL for IDS

Applying decision transformers and offline reinforcement learning to optimize intrusion detection systems. Focus on rule selection, false-positive reduction, and policy learning from historical security telemetry without online exploration risks.

Explainable AI for Certification

XAI techniques (SHAP, LIME, counterfactuals) for regulatory compliance and AI certification. Developing frameworks for auditable explanations in finance, healthcare, and safety-critical systems under GDPR, MiFID II, and AI Act requirements.

Multi-Agent Planning

Coordination and planning algorithms for multi-agent systems under uncertainty. Applications in autonomous vehicles, distributed cybersecurity defense, and collaborative robotics with theory-of-mind reasoning and trust models.

RAG Risk Agents

Retrieval-augmented generation for real-time risk management. Grounding LLM reasoning in policy documents, regulatory texts, and market data for explainable, auditable decision support in financial services.

Remote Sensing & Optimization

Cloud removal, image enhancement, and feature extraction from satellite imagery. Combining optimization, generative models, and explainability for Earth observation, agriculture, and environmental monitoring.

Privacy-Preserving AI

Differential privacy, federated learning, and secure multi-party computation for collaborative AI in healthcare, genomics, and cross-border data initiatives. Balancing utility with strong privacy guarantees.

Publications & Technical Reports

Continual Model-based Reinforcement Learning for Data Efficient Wireless Network Optimisation

Cengis Hasan et al. · Huawei Ireland Research Center

RL Networks

2023

Presents a continual RL approach achieving a 2× reduction in deployment lead-time for cell-level parameter optimisation in 5G networks, co-authored with Huawei Ireland Research Center.

UCB-driven Utility Function Search for Multi-objective Reinforcement Learning

Shi, Agapitos, Lynch, Cruciata, Cengis Hasan et al. · Huawei IRC & Trinity College Dublin

RL Networks

2023

Introduces a UCB acquisition method to efficiently explore scalarisation weight vectors for multi-objective RL, maximising Pareto front hypervolume on MuJoCo benchmarks.

LUCID-RAISE: Decision Transformer for Suricata IDS Rule Selection

Cognifinity S.à r.l-S · Luxembourg Ministry of Economy / NC3 / Luxinnovation

RL DT Cybersecurity

2024–2025

EU/Luxembourg-funded R&D (LU-CID Call Cybersecurity I) applying decision transformers to real-time Suricata rule management. Achieved 100% threat recall, 88% precision, and 94% F1 score. Technical report and formal specification available upon request.

Collaborate on Research

Interested in joint research, consortium participation, or technical consulting?

Get in Touch