Researcher at the University of the Faroe Islands. I survey, benchmark, and build ML systems for weather, maritime, and logistics, with GNN architecture expertise as the theoretical backbone.
My work consistently shows that data quality, evaluation design, and design-space understanding drive performance more than model complexity.
- Graph neural networks for physical simulation — design-space analysis of GNN architectures for fluid simulation (manuscript in preparation)
- Diagnostic weather benchmarking — conditional evaluation protocols that expose failure modes aggregate metrics hide, over complex maritime terrain (manuscript in preparation)
- Maritime data science — AIS vessel classification, adaptive transmission power, public datasets with RF metadata
- Logistics ML — GATv2 spatio-temporal encoders, transfer learning as integration middleware for low-data regimes
| Project | Description |
|---|---|
| AIS Vessel Classification | 174-configuration factorial experiment showing preprocessing explains 58% of performance variance (ICOIN 2026) |
| AIS-TSH Dataset | 7.8M AIS messages with per-message RF metadata (IEEE DataPort, DOI: 10.21227/fjhz-qf06) |
| Logistics Transfer Learning | GATv2 encoder with transfer learning for supply-chain graphs: 30–41% improvement, 4–10× data reduction (CITA 2026) |
| Adaptive AIS Transmission Power | Analytic collision-likelihood model: 35–64% collision reduction, 38–70% energy savings (ICSCCT 2026) |