Multimodal agent researcher spanning scientific figure benchmarks and efficient vision-language GUI grounding
2
papers here
2
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
—
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-09-20) 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 output metrics are available yet: the catalog holds only two 2026 preprints, so publication volume and productivity cannot be scored. This axis would measure counts of papers, publication venues, and output regularity over time.
Impact—
Impact cannot be scored yet: no citation counts, h-index, or downstream adoption data are present in the input. This axis would measure citations, influential citations, and uptake of the ReFigBench and RankGround artifacts.
Focus0/100
This is the weakest measured axis, scoring 0 on the 0-100 focus scale. The underlying field distribution is an exact 50/50 split between machine learning (multimodal agents and evaluation) and computer vision (GUI agents and vision-language models), computed from two 2026 catalog papers. The work is genuinely cross-cutting rather than scattered, but a two-paper base is too thin to call this a stable research identity yet.
Collaboration56/100
The collaboration axis scores 56 out of 100. Bibliometrics record 16 distinct coauthors across the reviewed work, with recurring names including Liyang Fan, Xinping Bi, Chi Wei, Guhong Chen, Chenghao Sun, Haoxiang Yang, Qingwen Li, Kai Yan, Hong Li, and Bo Li. On both 2026 papers Li sits in the middle author position rather than first or corresponding, a pattern consistent with team-based contribution inside larger groups.
Activity—
Activity is unscored: the data provide only two 2026 arXiv entries, with no multi-year publication history to establish a cadence. This axis would measure annual output rhythm, recency of contributions, and sustained research pace across years.
Reproducibility55/100
Reproducibility averages 55 out of 100 across 2 scored papers, and both Evolve-reviewed papers carry a medium reproducibility readiness rating. ReFigBench ships a 1,000-figure benchmark whose paired harness-versus-workflow study is described as broadly convincing, though with judge-family caveats; RankGround delivers strong, honestly analyzed results, with the +5.5% headline gain explicitly flagged for scrutiny.
Research summary
Yitai Li is an early-stage researcher working at the intersection of multimodal agents and vision-language models; this profile is name-resolved from public bibliographic data on two 2026 arXiv preprints reviewed in the Evolve catalog. The first, ReFigBench, introduces a 1,000-figure benchmark that frames scientific figure reconstruction as producing editable PowerPoint artifacts; Evolve's review finds a careful paired study showing that harness and workflow jointly decide agent quality, while flagging judge-family caveats around LLM-as-a-judge scoring. The second, RankGround, proposes efficient high-resolution GUI grounding via lightweight reranker-guided crop selection using a single VLM call; the review calls the results strong and honestly analyzed while noting the +5.5% headline gain needs scrutiny. The field distribution is an exact 50/50 split between machine learning (multimodal agents and evaluation) and computer vision (GUI agents and vision-language models), so the focus axis scores 0. Recorded bibliometrics include 16 distinct coauthors, a collaboration score of 56, and a reproducibility mean of 55 across 2 papers, both rated medium readiness. On both papers Li is a middle author and not the corresponding author, consistent with a contributor role inside larger teams. Output, impact, and activity cannot be scored from the data provided.
Collaboration pattern
With 16 distinct coauthors recorded and a collaboration score of 56, Li works in sizeable teams. The recurring coauthor list spans Liyang Fan, Xinping Bi, Chi Wei, Guhong Chen, Chenghao Sun, Haoxiang Yang, Qingwen Li, Kai Yan, Hong Li, and Bo Li. On both 2026 papers Li is a middle author and not corresponding, suggesting a contributor role inside larger multimodal-agent groups rather than solo or lead-author work.
Research keywords
Highlights
ReFigBench: a 1,000-figure benchmark for editable PPTX figure reconstruction
RankGround: efficient high-resolution GUI grounding with a single VLM call
Timeline
- 2026ReFigBench preprint posted to arXiv: a 1,000-figure benchmark framing scientific figure reconstruction as editable PowerPoint artifacts, with a paired harness-versus-workflow agent study.
- 2026RankGround preprint posted to arXiv: lightweight reranker-guided crop selection for efficient high-resolution GUI grounding via a single VLM call.
- 2026Both catalog papers appear with middle, non-corresponding authorship across an even machine learning / computer vision split, alongside 16 distinct coauthors.
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 and radar scores come from the platform's computation on public bibliographic data only; no institution, country, or ORCID was available to verify identity.
- Output, impact, and activity dimensions are unscored (dimMissing): no citation counts, h-index, or multi-year publication history were provided.
- Only two 2026 arXiv preprints were reviewed, both with medium reproducibility readiness, so conclusions rest on a small and very recent base.
- The frequent-coauthor list contains one duplicate name and the profile name itself; these were treated as data noise, and name resolution relies on bibliographic matching only.