ML-interpretability and biosecurity researcher: foundation-model mechanism analysis and AI bio-risk governance
2
papers here
2
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
—
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-15) and cover the author's entire public record, not just the papers listed here. Reproducibility signal comes from Evolve's own reviews.
Research profile
Output—
Publication-output metrics — total papers, venue quality, and first- or corresponding-author volume — are not yet available; only two arXiv preprints have been reviewed in the Evolve catalog, too small a base to score productivity.
Impact—
Citation counts, h-index, and downstream uptake are absent from the current data; this axis would measure how widely the 2026 preprints and any earlier work have influenced interpretability and biosecurity research.
Focus0/100
The focus score is 0/100, reflecting an evenly split field distribution: 0.5 for computational biology / machine-learning interpretability and 0.5 for biosecurity / AI governance at the computational biology interface. The two cataloged 2026 preprints sit one in each field — a mechanism-oriented interpretability census and a governance-oriented literature review — indicating a deliberately dual-facing agenda rather than concentration in a single line of work; the low score measures dispersion, not weakness.
Collaboration28/100
The collaboration score is 28/100, based on 3 distinct coauthors recorded across the catalog: Aris Karatzikos, Alexandros Tzanakakis, and Candace S. Y. Chan. Both 2026 preprints place the researcher in the last author position, though neither credits a corresponding role, a pattern consistent with a small, stable team rather than a large distributed consortium; the compact network also means most bibliographic weight rests on a single core group.
Activity—
Activity metrics — publication cadence and recency across recent years — are insufficient; the reviewed record holds only two papers, both dated 2026, so no year-over-year trend can be computed yet.
Reproducibility60/100
Reproducibility scores 60/100, the mean across the 2 papers assessed. The split is stark: the gated-FFN interpretability census carries high reproducibility readiness, consistent with Evolve's description of a careful census built on causal ablation, while the AI-and-biosecurity literature review rates low, as reviews rest on synthesizing unreviewed preprints rather than releasing artifacts. The average therefore describes a mixed, artifact-dependent profile.
Research summary
Ilias Georgakopoulos-Soares (ORCID 0000-0003-3641-1488) is an early-stage researcher working at the interface of machine learning and the life sciences. The Evolve catalog reviews two 2026 arXiv preprints, and on both the researcher holds the last author position without the corresponding role. The first paper, 'Structure is not mechanism,' reports a census of high-gain gated-FFN rows across text and genomic foundation models, including DNABERT-2, connecting super weights and spectral concentration through causal ablation; Evolve judges it careful and unusually honest, with high reproducibility readiness. The second, on artificial intelligence and biosecurity, maps capability, threat pathways, and a defense-in-depth governance framework; Evolve finds the framework sensible but notes heavy reliance on unreviewed preprints, rating reproducibility readiness low. The field distribution splits evenly (0.5/0.5) between computational biology / machine-learning interpretability and biosecurity / AI governance. The recorded network is compact — 3 distinct coauthors — giving a collaboration score of 28/100, while reproducibility averages 60/100 across the two assessed papers. Output, impact, and activity metrics are unavailable, so this profile should be read as an early-stage, name-resolved bibliographic sketch.
Collaboration pattern
The recorded collaboration footprint is compact: 3 distinct coauthors — Aris Karatzikos, Alexandros Tzanakakis, and Candace S. Y. Chan — underlie a collaboration score of 28/100. On both 2026 cataloged preprints the researcher sits in the last author position without the corresponding designation, suggesting a senior role within a small, recurring core team rather than broad cross-institutional work. These metrics come only from two reviewed papers, so the network may expand as more of the record is indexed.
Research keywords
Highlights
Structure is not mechanism: a cross-model census of high-gain gated-FFN rows
Defense-in-depth governance for AI biosecurity risk
Timeline
- 2026Last-author preprint on high-gain gated-FFN rows across text and genomic foundation models, rated high reproducibility readiness by Evolve.
- 2026Last-author literature review on artificial intelligence and biosecurity proposing a defense-in-depth governance framework, rated low reproducibility readiness.
- 2026Both cataloged outputs appear in the same year, splitting the field distribution evenly between ML interpretability and biosecurity / AI governance.
Papers in the Evolve catalog
Frequent co-authors
About this profile
This profile is AI-structured by Evolve.Pub from public bibliographic data (OpenAlex) and Evolve's own paper reviews. It is not written or endorsed by the researcher unless the profile is claimed. Radar dimensions are computed from real bibliometrics with fixed formulas — raw figures are shown alongside; missing dimensions mean insufficient data, never zero. Errors? Claim the profile to correct or hide it.
- Bibliometric scores are computed from a limited OpenAlex-based snapshot covering only two reviewed papers; global citation and output figures are absent.
- Institution and country are not confirmed in the data; the profile is name-resolved via ORCID 0000-0003-3641-1488 from public bibliographic records.
- Both cataloged papers are 2026 arXiv preprints; peer-review status and final venues are unverified.
- Output, impact, and activity radar axes lack sufficient underlying data and are left unscored.