Evolve.Pub
Sign inScan your paper
TG

Tao Gui

Fudan University

Unclaimed · AI-structured profileORCID OpenAlex

Tao Gui — multimodal LLM post-training and evidence-grounded LLM evaluation at Fudan University

5

papers here

4

reviewed by Evolve

0

first-author

0

corresponding

—

works · OpenAlex

—

h-index · OpenAlex

Bibliometric figures come from OpenAlex (2026-10-17) 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: insufficient dataOutputImpact: insufficient dataImpactFocus: 0/100FocusCollaboration: 83/100CollaborationActivity: insufficient dataActivityReproducibility: 66/100Reproducibility

Output—

Not yet scored. This axis would quantify total publication volume, venue quality and productivity trend over time; the current snapshot lists only the four 2026 arXiv papers reviewed by Evolve, which is too little to compute it.

Impact—

Not yet scored. This axis would measure citation-based influence — citations, h-index and downstream uptake of the work — but no citation statistics were provided in the input data, so no estimate is offered here.

Focus0/100

The focus score of 0/100 reflects an unusually even spread in the raw field distribution: multimodal LLM post-training, scientific reasoning benchmarks, LLM evaluation for peer review, and automated LLM fine-tuning each account for exactly 0.25 of the four catalog papers. The current record shows breadth-first work across adjacent LLM training and evaluation topics rather than convergence on one narrow specialty.

Collaboration83/100

With a collaboration score of 79/100 built on 51 distinct coauthors across only four cataloged papers, Tao Gui works in dense team settings: every paper carries multiple coauthors, and the frequent-coauthor list is dominated by a recurring Fudan-centered core — Xuanjing Huang, Qi Zhang, Zhiheng Xi, Enyu Zhou, Jiazheng Zhang, Dingwei Zhu, Haoran Zhao, Wei Du, Dingwen Yang, Jixuan Huang and Junlin Shang. Authorship sits mostly in middle positions, with one last-author role on Sci-MMR.

Activity—

Not yet scored. This axis would track publishing cadence — papers per year and recency of output — but the snapshot contains only four 2026 preprints, which is insufficient to establish a stable tempo.

Reproducibility66/100

The reproducibility score of 66/100 is the mean readiness rating across 4 reviewed papers (reproMean 66, reproPapers 4). Evolve rates NovGauge as high readiness, while ReDraft, Sci-MMR and SAGE are medium: the benchmark work ships auditable artifacts, but several headline claims carry unresolved caveats, such as SAGE's n=3 runs without significance tests and NovGauge's reliance on a single LLM judge.

Research summary

Tao Gui is a researcher at Fudan University whose recent work, as captured in Evolve's review catalog, concentrates on training and evaluating large language and vision-language models. Four 2026 arXiv papers define the current record, splitting evenly across four directions: continual post-training for vision-language models (ReDraft), multi-step evidence-grounded scientific reasoning benchmarks for multimodal agents (Sci-MMR), fine-grained LLM evaluation of paper novelty for peer review (NovGauge), and automated LLM fine-tuning with accumulated strategies (SAGE). The pattern reads as someone who both builds post-training methods and constructs the benchmarks needed to audit them. Authorship positions — middle on three papers, last on Sci-MMR, none flagged corresponding — fit contributor roles inside a large team: 51 distinct coauthors appear across only four papers, and the frequent-coauthor list centers on a recurring Fudan group including Xuanjing Huang, Qi Zhang and Zhiheng Xi. Evolve's verdicts are broadly positive but precise: Sci-MMR is called genuinely diagnostic and unusually well-audited (235 tasks, LLM-judged coverage); SAGE lifts Avg@3 from 3.2% to 15.6% on nine tasks, though with n=3 runs and no significance tests; NovGauge is soundly built but hinges on a single LLM judge; ReDraft shows strong results with a slight clash against its own Table 4. Reproducibility readiness averages 66 across the four papers, one rated high and three medium.

Collaboration pattern

Across four cataloged 2026 papers, 51 distinct coauthors appear, supporting a collaboration score of 79/100. The work is team-based within a stable Fudan-centered circle: frequent coauthors include Xuanjing Huang, Qi Zhang, Zhiheng Xi, Enyu Zhou, Jiazheng Zhang, Dingwei Zhu, Haoran Zhao, Wei Du, Dingwen Yang, Jixuan Huang and Junlin Shang. Tao Gui holds middle-author positions on three papers and the last-author position on Sci-MMR; no paper is flagged as corresponding author.

Research keywords

Vision-language model post-trainingContinual learning and catastrophic forgettingSelf-revision and on-policy distillationScientific reasoning benchmarksArgument graphsEvidence groundingMulti-hop reasoningLLM evaluation for peer reviewNovelty assessmentAutomated LLM fine-tuningMonte Carlo Tree SearchMulti-agent experience reuse

Highlights

Timeline

  1. 2026Co-authored NovGauge, a fine-grained benchmark diagnosing LLMs' capability in paper novelty assessment, in a middle-author role.
  2. 2026Appeared as last author on Sci-MMR, benchmarking multi-step evidence-grounded scientific reasoning in multimodal agents.
  3. 2026Contributed to ReDraft on reference-driven revision for continual vision-language model post-training.
  4. 2026Contributed to SAGE, strategy accumulation and guided execution for automated LLM fine-tuning.

Papers in the Evolve catalog

Frequent co-authors

Xuanjing Huang×3Shaofan Liu×2Honglin Guo×2Qi Zhang×2Zhiheng Xi×2Enyu Zhou×2Jiazheng Zhang×2Dingwei Zhu×2Qi Zhang×2Xuanjing Huang×2Shihan Dou×1Haoxiang Jia×1

Is this you? Claim this profile

Enter the email you published with. We'll send a one-time claim link; after claiming you can edit the introduction or hide this page. Requests from a mismatched address go to manual review.

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 public bibliometric data only; this snapshot includes no citation counts, h-index or full publication list.
  • Coverage is limited to the four 2026 arXiv papers reviewed by Evolve; other work by the same author may exist outside this catalog.
  • The profile is name-resolved from public bibliographic data anchored to ORCID 0000-0002-0059-0210 and the Fudan University affiliation; 'Tao Gui' is a relatively common Chinese name.
  • The output, impact and activity axes are marked missing because the underlying metrics were not provided, and all catalog papers are preprints without recorded venue acceptance.
Tao Gui — Researcher Profile | Evolve.Pub