Taylor T. Johnson

Vanderbilt University

Unclaimed · AI-structured profileOpenAlex

Formal verification of ML at Vanderbilt: Star-set reachability, GNN verification for power systems, NNV3 toolchain.

papers here

reviewed by Evolve

first-author

corresponding

—

works · OpenAlex

—

h-index · OpenAlex

Bibliometric figures come from OpenAlex (2026-10-02) 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—

Not scored: the input snapshot contains no publication-count, venue, or productivity metrics for this axis. It would measure total research output — number of publications over time and their venues — once fuller bibliographic data is ingested.

Impact—

Not scored: no citation counts, h-index, or downstream-usage data are available in the provided snapshot. This axis would measure the influence of the work — citations and uptake of tools like NNV3 by the verification community — and remains unscored.

Focus0/100

Scores 0/100, derived from a field distribution split exactly 50/50 across two labels: formal verification of machine learning (computer science / cyber-physical systems) and machine learning formal verification (GNN verification for power systems). Both labels describe the same core area, so the zero reflects an even split across field labels rather than scattered interests; the cataloged work is topically concentrated on neural-network verification.

Collaboration55/100

Scores 55/100 from real bibliometric data: 15 distinct coauthors appear in the snapshot, with 12 recurring frequent collaborators including Anne M. Tumlin, Ben Wooding, Diego Manzanas Lopez, Samuel Sasaki, Navid Hashemi, Waseem Abbas, and Tyler Derr. On both cataloged 2026 papers Johnson sits in the last author position without being corresponding author, indicating a senior, supervisory collaboration pattern within a stable group.

Activity—

Not scored: the snapshot lacks year-by-year publication frequency data, so recent publishing cadence cannot be quantified. This axis would measure ongoing research activity — outputs per year over time — though both cataloged papers are dated 2026.

Reproducibility100/100

A perfect 100/100, computed as the mean reproducibility-readiness score across 2 papers. Both reviewed outputs — the GraphStar-based GNN verification paper and the NNV3 tool paper — are rated high readiness. This matches the tool-building character of the work: NNV3 ships as a unified MATLAB verification tool, suggesting methods are released in a form others can re-run.

Research summary

Taylor T. Johnson is a researcher at Vanderbilt University working at the intersection of formal verification and machine learning, with a cataloged focus on neural-network verification via reachability analysis and Star sets. Evolve's review covers two 2026 arXiv preprints. The first, on reachability-based formal verification of graph neural networks with node and edge features, claims the first edge-aware formal verification of GNNs through GraphStar sets; Evolve judged the method sound, with a strong comparison against CORA, while flagging issues in PF feature design and selection. The second, NNV3, unifies weight-perturbation, video, GNN, probabilistic, and fairness verification in a single MATLAB tool; the integration is judged credible, though evaluations remain small. Johnson appears as last author on both papers — typically the senior position — without being listed as corresponding author. The bibliometric snapshot records 15 distinct coauthors and 12 frequent collaborators, a collaboration score of 55/100, and a field distribution split evenly between formal verification of machine learning and GNN verification for power systems — effectively two labels for one coherent verification agenda. Both reviewed papers rate high on reproducibility readiness, producing a mean score of 100/100 across 2 papers. Output, impact, and activity are left unscored because the underlying data is unavailable. The overall picture is a tool-oriented verification researcher building reusable formal methods for learning-enabled systems.

Collaboration pattern

The snapshot records 15 distinct coauthors, with 12 recurring frequent collaborators — including Anne M. Tumlin, Ben Wooding, Diego Manzanas Lopez, Samuel Sasaki, Navid Hashemi, Waseem Abbas, Ipek Oguz, Meiyi Ma, and Tyler Derr — and a collaboration score of 55/100. On both cataloged papers Johnson occupies the last author position without being corresponding author, a pattern consistent with senior supervision of a rotating group spanning GNN verification, power systems, and cyber-physical applications.

Research keywords

neural network verificationformal verification of machine learninggraph neural networksreachability analysisStar setsadversarial robustnessweight perturbationGNN verification for power systemscyber-physical systemsfairness verificationprobabilistic verificationverification tooling in MATLAB

Highlights

First edge-aware formal verification of GNNs via GraphStar sets
Evolve's review of this 2026 paper (Johnson as last author) notes its claim of the first edge-aware formal verification of graph neural networks using GraphStar sets. The method is judged sound with a strong comparison against CORA, while PF feature design and selection are flagged as open issues.
NNV3 unifies five verification tasks in one MATLAB tool
The NNV3 paper expands neural network verification to new architectures and domains, unifying weight-perturbation, video, GNN, probabilistic, and fairness verification in a single MATLAB tool. Evolve finds the integration credible, though the evaluations remain small in scale.
Perfect reproducibility readiness across reviewed outputs
Both 2026 cataloged papers are rated high on reproducibility readiness, yielding a mean score of 100/100 across 2 papers — consistent with the tool-building orientation of the research, where verification methods are packaged for others to run.

Timeline

2026Published reachability-based formal verification of graph neural networks with node and edge features, introducing GraphStar sets (arXiv, last author).
2026Released NNV3, expanding neural network verification to weight-perturbation, video, GNN, probabilistic, and fairness domains within one MATLAB tool (arXiv, last author).
2026Both cataloged papers rated high on reproducibility readiness, producing a 100/100 mean score across 2 reviewed papers.

Papers in the Evolve catalog

Frequent co-authors

Anne M. Tumlin×2 Ben Wooding×2 Diego Manzanas Lopez×2 Samuel Sasaki×1 Muhammad Usama Zubair×1 Navid Hashemi×1 Hongchao Zhang×1 Waseem Abbas×1 Ipek Oguz×1 Meiyi Ma×1 Zhenxuan Shao×1 Tyler Derr×1

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 signals come from an OpenAlex-based public snapshot; no citation counts, h-index, or venue-level metrics are included in the input data.
Output, impact, and activity dimensions are unscored because the underlying data is insufficient; no estimates were made.
Only the two 2026 arXiv preprints reviewed by Evolve are covered; the profile may not reflect the researcher's full publication record.
Identity is name-resolved from public bibliographic data (Taylor T. Johnson, Vanderbilt University); no ORCID was provided for disambiguation.