Digital Topology Instability under Exact Decimation of Retinal Vessel Annotations: Stage Localization and Source Provenance
Accepted, to appear
Lucian Murgu
Software Engineer · Applied AI & Research
Hi, I’m Lucian. I build software and study measurement. When a score says a system works, I want to know what it isn’t telling us.
A branching network grown in your browser — the kind of thin, connected structure my research measures.Point at a branch to trace its path to the root.
The thread
I started in structural engineering, working on bridge-design software at Bentley Systems. When a calculation looked wrong and the debugger couldn’t help, I recomputed it by hand until the number and the structure agreed.
Since 2018 I have built software for European institutions — most recently the system independent experts use to score grant proposals, where an evaluation result has real consequences.
Now, in my PhD at the Romanian Academy and in my own practice, I ask the same question of AI systems: does the score, the explanation or the agent’s “done” actually reflect what is underneath?
Does the number tell the truth about the structure behind it?
Selected work
Each project asks that question of a different system.
Same Dice, Different Geometry
Does a high overlap score mean the structure is right?
How matched Dice scores can conceal different structural errors in controlled vessel-mask surrogates.
My role First author, with Tudor Barbu (Romanian Academy).
AgroUnu
When a coding agent says “done”, is the task actually done?
Coding-agent evaluation through bounded tasks, recorded trajectories and explicit acceptance checks.
My role I design and build the platform and the agent-evaluation process around it.
Structural mechanics / numerical evaluation
Does an elegant geometry predict behaviour better than a simple baseline? It didn’t — and that result is reported.
Comparing behavioral distances in synthetic spring networks, including a negative result for the tested local geometry.
My role I designed, preregistered and ran the experiment, and wrote the report.
Mechanistic interpretability
Is a model right for the right reasons?
An ongoing research direction in understanding how neural networks represent and process information.
My role My own research direction, alongside the PhD.
Research interests →Try it: right prediction, wrong reason →
Independent research
Projects I set up on my own, run with the same discipline — including the results that said no.
RSKAN — interpretable Kolmogorov–Arnold networks
Can a network that refines its own function state recover the underlying law better than classical methods?
A preregistered research prototype with decision gates fixed before each run. The first two chapters ended in documented negative results — sparse regression did it more simply. In a later multi-tone signal family the learned model beat the best classical baseline on every seed, at about 43% of its compute — within three-tone cases only; it did not transfer across model order.
Virtual Cell Challenge 2026
How do human cells’ gene-expression profiles respond to genetic perturbations?
A competition entry predicting perturbation responses for 300 target genes from public Perturb-seq data, built as a hash-locked, independently verifiable pipeline. The final entry is submitted; scores are not yet public.
Now
- Second year of my PhD on fractal geometry and topology in medical image analysis; a new paper accepted at IEEE EHB 2026.
- Building evaluation and monitoring tools for coding agents.
- Open to research-engineering roles, and to selected consulting engagements.
Experience
- Independent practiceSoftware & AI consultantFeb 2026–present
- European Commission · DG RTDExternal ICT ContractorNov 2019–Oct 2026
- Bentley SystemsSoftware Developer and Structural EngineerFeb 2015–Apr 2017
Recent publications
Multi-Task U-Net Architecture for Tumor Segmentation and Recognition based on Fractal Analysis and Nonlinear Reaction-Diffusion
Let’s talk
If you are working on a question like these — an evaluation you are not sure you can trust, a model whose reasons matter, or a role where this kind of care is the job — I would like to hear from you.