Research
Research.
I investigate how we evaluate, understand and build reliable AI systems, and how geometric and numerical methods can be used in medical image analysis.
My work combines AI evaluation and interpretability, agentic systems, geometric modeling and scientific computing, through controlled experiments and reproducible studies.
Research areas
- AI Evaluation & ReliabilityMetrics, failure analysis, robustness
- Medical Image AnalysisSegmentation, topology, vessel analysis
- Geometric & Numerical MethodsFractals, multiscale analysis, structural modeling
- Experimental Machine LearningInterpretable models, comparisons, ablations
Selected investigations
Featured · Evaluation methodology
Same Dice, Different Geometry
How identical Dice scores can conceal different geometric structure and connectivity in vessel segmentation, shown with 300 controlled surrogates of expert retinal vessel annotations.
Medical image analysis
Digital Topology Instability under Exact Decimation
A systematic study of how exact downsampling alters vessel-mask connectivity and topology before skeleton or graph extraction.
Numerical methods
Structural Mechanics / Numerical Evaluation
A preregistered comparison of geometric distances for structural systems, reporting a negative result against a stronger baseline.
Experimental machine learning
RSKAN: Interpretable Kolmogorov–Arnold Networks
Preregistered experiments with recursive, interpretable KAN-style models, compared against the strongest classical baselines.
f = 1
f = 2
f = 4No investigations under this topic.
More researchAll investigations, publications and competition entriesFigures are synthetic illustrations and schematics drawn for this page, with computed values; study data appear in each linked report.