Fudan University researcher bridging scientific knowledge infrastructure and LLM agent security
2
papers here
2
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
—
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-01) 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—
No publication-volume data is available yet. This axis would measure total scholarly output — paper counts, venues, and yearly productivity — so it is left unscored rather than estimated.
Impact—
Citation-based influence data is not yet available for this profile. This axis would measure citations, citing works, and field-weighted impact; it remains unscored rather than approximated.
Focus0/100
The focus score is 0, driven directly by an even 50/50 field distribution: half of the reviewed output sits in scientific information retrieval / AI for Science (knowledge infrastructure) and half in AI safety / LLM agent security. Rather than deep specialization in one lane, the cataloged work bridges two distinct research communities, so topical concentration is currently minimal.
Collaboration68/100
Collaboration scores 68 out of 100, backed by 29 distinct coauthors across the reviewed record and a stable core of 12 recurring collaborators, including Yuan Huang, Sihan Hu, and Hongyu Gu. Both cataloged papers list Yu Li in a middle author position without corresponding status, consistent with contributing inside larger Fudan-affiliated teams rather than solo-led efforts.
Activity—
Year-by-year activity data is insufficient. This axis would measure publishing cadence and continuity over time; with only 2026 records in scope, it stays unscored.
Reproducibility60/100
Reproducibility averages 60 across the two papers with recorded readiness checks (reproPapers = 2, reproMean = 60). The split is stark: ClashBench is rated high readiness as an executable benchmark, while Large Knowledge Model is rated low, with small untested gains and unmeasured graph fidelity. The composite score therefore masks a real transparency gap between the two projects.
Research summary
Yu Li is a researcher at Fudan University whose reviewed work to date spans two distinct areas in roughly equal measure: scientific information retrieval and AI for Science (knowledge infrastructure) on one side, and AI safety / LLM agent security on the other, each accounting for 50% of the cataloged field distribution. Two 2026 arXiv papers anchor this early-stage profile. The first, Large Knowledge Model, converts more than 40 million papers into reasoning graphs and reports question-answering gains of up to 14.69 percentage points; Evolve's verdict is cautiously positive but flags small, untested margins and unmeasured graph fidelity, with low reproducibility readiness. The second, ClashBench, is an executable agent-safety benchmark showing that LLM agents destructively preempt incumbent tasks in 44.5% of runs — rated high for reproducibility, though single-run statistics warrant caution. Collaboration data shows 29 distinct coauthors and a recurring core of 12 collaborators, with a collaboration score of 68; both papers carry middle author positions without corresponding status. Output, impact, and activity metrics are not yet available and are left unscored. Overall this is a dual-track, team-based early profile whose reproducibility record is promising but uneven.
Collaboration pattern
The collaboration score of 68 rests on 29 distinct coauthors and a recurring core of 12 frequent collaborators — including Yuan Huang, Sihan Hu, Hongyu Gu, Chao Ma, and Jiaxing Zhang. Both cataloged 2026 papers place Yu Li in a middle author position without corresponding-author status, suggesting contribution within larger Fudan-affiliated teams spanning both the knowledge-infrastructure and agent-security tracks rather than solo-led projects.
Research keywords
Highlights
Large Knowledge Model: 40M+ papers turned into reasoning graphs
ClashBench: measuring destructive preemption by LLM agents
A two-track agenda split exactly 50/50
The reviewed field distribution splits evenly between scientific knowledge infrastructure and LLM agent security, with both 2026 arXiv papers carrying middle author positions — an early-stage, deliberately dual-domain research program.
Timeline
- 2026First cataloged paper, Large Knowledge Model, appears on arXiv, framing a scientific reasoning landscape over 40M+ papers with question-answering gains of up to 14.69 percentage points.
- 2026Second cataloged paper, ClashBench, appears on arXiv, introducing an executable agent-safety benchmark with a 44.5% destructive preemption rate across runs.
- 2026Reviewed output splits evenly across two fields — knowledge infrastructure and agent security — each at 50% of the cataloged distribution, both in middle author positions.
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.
- Radar scores are computed from real bibliometric data, but only for collaboration, focus, and reproducibility; output, impact, and activity remain unscored.
- Qualitative assessments draw on the two Evolve-reviewed 2026 arXiv preprints only; no peer-reviewed venue record is included in this profile.
- 'Yu Li' is a common name and no ORCID is available; this profile is name-resolved from public bibliographic data and may merge or omit records.
- The reproducibility mean (60) is based on only two papers, so the score is indicative rather than stable.