ML security researcher at University of Kentucky spanning gradient inversion attacks and agentic-AI anomaly detection
2
papers here
2
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
8
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-16) 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 yet scored: verified venue-level publication data is unavailable in this snapshot, so total productivity cannot be assessed. This axis would measure the volume, venues, and forms (journal, conference, preprint) of the researcher's works over time.
Impact—
Not yet scored: full citation-linked data is unavailable in this snapshot, so influence cannot be rated. This axis would measure citation-based influence — aggregations beyond the reported h-index of 8 and i10-index of 6, per-work citations, and venue-level attention.
Focus0/100
The focus score is 0, reflecting an even 50/50 field split between machine learning security and privacy (gradient inversion attacks) and AI security (agentic systems, anomaly detection). Topic coverage is correspondingly broad: adversarial robustness (16), privacy-preserving technologies (13), cryptography and data security (10), security and verification (10), and blockchain (7).
Collaboration51/100
The collaboration score is 51, based on 12 distinct coauthors in the snapshot. Recurring partners include Wenjing Lou, Y. Thomas Hou, Yulia R. Gel, Ning Zhang, Yang Xiao, and Yi Shi. Both Evolve-reviewed 2026 papers list the author in a middle position without a corresponding role, indicating contribution within multi-author teams rather than solo-led work.
Activity98/100
The activity score is 98, the strongest axis of this profile. OpenAlex yearly counts climb from 1 work in each of 2020–2022 to 4 (2023), 5 (2024), 8 (2025), and 15 (2026), with 28 works in the most recent three years — a steep and ongoing acceleration in research tempo.
Reproducibility38/100
Reproducibility scores 38, a mean across two Evolve-assessed papers. The temporal gradient inversion preprint is rated low readiness, while Skynet is medium: its 100%-recall / 0.89%-FPR results come from single runs with small anomaly sets and a domain-shift confound, limiting current replication confidence.
Research summary
Shanghao Shi is a machine learning security researcher at the University of Kentucky whose OpenAlex-resolved record shows a steep output ramp: yearly counts rise from single works in 2020–2022 to 4 in 2023, 5 in 2024, 8 in 2025, and 15 in 2026 — 35 works in total, 28 of them in the last three years. Citation impact remains early-stage, with an h-index of 8 and an i10-index of 6. The record splits evenly, 50/50, between machine learning security and privacy (gradient inversion attacks) and AI security (agentic systems, anomaly detection); leading topics are adversarial robustness (16), privacy-preserving technologies (13), cryptography and data security (10), and security and verification in computing (10). Two Evolve-reviewed 2026 arXiv preprints exemplify both lines: a temporal gradient inversion attack reconstructing embodied reinforcement learning trajectories at 18.8 dB PSNR with 100% action accuracy, and Skynet, a benign-only workflow-level anomaly detector for agentic AI reporting 100% recall at a 0.89% false positive rate. Reproducibility evidence is thin: mean readiness 38 across just two assessed papers.
Collaboration pattern
The profile shows a moderate collaboration footprint — 12 distinct coauthors — with a stable recurring circle including Wenjing Lou, Y. Thomas Hou, Yulia R. Gel, Ning Zhang, Yang Xiao, Yi Shi, Sudip Bhujel, Ruiquan Huang, and Chaoyu Zhang. Both Evolve-reviewed 2026 papers place Shanghao Shi in a middle authorship position without a corresponding role, suggesting research conducted within multi-author teams rather than solo-led projects.
Research keywords
Highlights
Temporal gradient inversion reconstructs embodied RL trajectories
Skynet: benign-only workflow-level anomaly detection for agentic AI
Output acceleration across two security research lines
OpenAlex yearly counts rise from 1 work per year in 2020–2022 to 15 in 2026, with 28 works in the last three years, split evenly between gradient-inversion privacy attacks and agentic-AI anomaly detection.
Timeline
- 2020Publication record opens with 1 work in the OpenAlex yearly counts.
- 2022The early phase closes at a pace of 1 work per year (2020–2022).
- 2023Output rises to 4 works, the first step up from single annual publications.
- 20245 works recorded, continuing the upward trajectory.
- 20258 works recorded in a year of accelerating output.
- 202615 works recorded, including both Evolve-reviewed arXiv preprints: the temporal gradient inversion study and Skynet's agentic-AI anomaly detection.
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.
- Bibliometrics (counts by year, h-index, i10-index, topics, distinct coauthors) come from an OpenAlex-resolved identity; name ambiguity is possible for common names.
- Output and impact radar axes are unscored because the underlying data is insufficient in this snapshot.
- The frequent-coauthor list contains "Ning Zhang" twice, which may denote two different researchers sharing a name.
- Reproducibility rests on only two Evolve-assessed 2026 arXiv preprints (mean readiness 38); both are middle-authorship, non-corresponding works.