ML evaluation · Geospatial AI · Visual Attention · Robustness & Auditability
Final-year CS at Cardiff University. I build and interrogate ML systems — with a focus on what models get wrong, why, and how to know when to trust them.
- ML evaluation and robustness — identifying hidden model failure, shortcut reliance, and unreliable performance signals
- Geospatial AI — remote sensing, segmentation, and input-efficiency for landslide detection
- Visual attention modelling — analysing where saliency models diverge from human behaviour
- Explainability and auditability — feature selection, attribution, and evidence for trustworthy model use
| Project | Description | Status |
|---|---|---|
| Landslide Segmentation | SFFS channel selection for multi-spectral landslide detection | Under review |
| Saliency Evaluation | Hidden failure detection in visual attention models | Active |
Ahmad A., Karakuş O., Rosin P.L. (2025).
Sequential Feature Selection for Efficient Landslide
Segmentation from Multi-Spectral Data.
Frontiers in Remote Sensing — under review.
