Yuan Wu

Jilin University · United States

Unclaimed · AI-structured profileOpenAlex

Jilin University researcher building benchmarks that stress-test LLM and multimodal models on critique and reasoning

papers here

reviewed by Evolve

first-author

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—

Not yet scored: citation-normalized publication metrics are unavailable in this snapshot (OpenAlex records h-index and i10-index of 0). This axis would measure publication volume, venue quality, and output consistency over time.

Impact—

Insufficient data: no citation counts, h-index trajectory, or influence signals are available yet. This axis would measure citation uptake, impact on follow-up work, and recognition within the benchmarking and NLP communities.

Focus—

Not scored: OpenAlex topic shares are recorded, but the composite focus measure has not been computed. This axis would measure how concentrated the research agenda is across topics versus how widely it disperses.

Collaboration36/100

Collaboration scores 36/100, reflecting a compact, team-centric pattern: the OpenAlex snapshot lists only 5 distinct coauthors, all recurring — Yi Chang, Gengxu Li, Zhongru Chen, Jinzhe Li, and Jinnan Li. Across the three 2026 catalog papers Yuan Wu sits in a middle authorship position and is not the corresponding author, suggesting contribution within an established group rather than an independent or externally networked footprint.

Activity87/100

Activity scores 87/100, the strongest signal in this profile: OpenAlex attributes 19 works to the last three years, with the 2026 count alone at 19, and the Evolve catalog adds three 2026 arXiv preprints (RPCBench, OCR-MetaReasoning Benchmark, MMPCBench). Topic counts — 13 in topic modeling, 7 in NLP techniques, 6 in multimodal applications — confirm a high-frequency pace concentrated in a single productive year.

Reproducibility—

No data yet: this axis would aggregate artifact release, code and data availability, and reproducibility readiness into a profile-level score; the two reviewed benchmarks are each marked medium readiness, but no overall rating exists.

Research summary

Yuan Wu is a researcher at Jilin University whose work centers on evaluation benchmarks for large language models and multimodal large language models (MLLMs). The OpenAlex snapshot shows a highly concentrated output pattern: 19 works in the last three years, all recorded in 2026, driving an activity score of 87/100. Three 2026 arXiv preprints in the Evolve catalog define the agenda. RPCBench provides a 4,623-instance benchmark for proactive premise critique in LLM-based recommendation; Evolve credits its taxonomy and metrics while flagging LLM generator-judge circularity. The OCR-MetaReasoning Benchmark evaluates meta-reasoning across 1,500 text-rich image samples, finding deduction and layout hardest for MLLMs. MMPCBench extends proactive critique to flawed multimodal inputs and is not yet reviewed. Topic counts — 13 in topic modeling, 7 in NLP techniques, 6 in multimodal applications, 5 in domain adaptation and few-shot learning, 3 in handwritten text recognition — indicate breadth beyond benchmarking. Collaboration is compact: 5 distinct coauthors (Yi Chang, Gengxu Li, Zhongru Chen, Jinzhe Li, Jinnan Li), with middle rather than corresponding authorship on catalog papers and a collaboration score of 36/100. Citation-based output and impact cannot yet be scored (h-index and i10-index are 0 in the snapshot), so this reads as an early-stage but exceptionally active benchmarking profile.

Collaboration pattern

Collaboration follows a compact, team-based pattern. The OpenAlex snapshot records 5 distinct coauthors, and the same five names recur: Yi Chang, Gengxu Li, Zhongru Chen, Jinzhe Li, and Jinnan Li. Across the three 2026 catalog papers Yuan Wu holds a middle authorship position and is not the corresponding author, consistent with contributing within an established group — a collaboration score of 36/100 reflects this narrow but steady circle.

Research keywords

LLM-based recommendationBenchmark constructionProactive premise critiqueLLM-as-a-judgeOCR reasoningText-rich image understandingMultimodal large language modelsMeta-reasoning evaluationTopic modelingHandwritten text recognitionDomain adaptation and few-shot learning

Highlights

RPCBench: 4,623 instances for proactive premise critique in LLM recommendation
Evolve's review credits this benchmark's taxonomy and metrics for flagging faulty premises in LLM-based recommendation, while cautioning that using LLMs as both generator and judge creates circularity that weakens the results. Reproducibility readiness: medium.
OCR-MetaReasoning: probing MLLM meta-reasoning on text-rich images
A rigorous 1,500-sample benchmark finds deduction and layout the hardest skills for multimodal large language models in text-rich image understanding; Evolve notes that data contamination and abductive-only edits limit comparisons across reasoning types. Reproducibility readiness: medium.
MMPCBench: extending proactive critique to flawed multimodal inputs
A third 2026 benchmark stress-tests multimodal large language models on critiquing flawed inputs, rounding out a proactive-critique line spanning recommendation, OCR reasoning, and multimodal understanding. Evolve has not yet published a review.

Timeline

2026Published RPCBench, a 4,623-instance benchmark for proactive premise critique in LLM-based recommendation; Evolve rates its taxonomy and metrics valuable while noting generator-judge circularity.
2026Released the OCR-MetaReasoning Benchmark, evaluating MLLM meta-reasoning on 1,500 text-rich image samples; deduction and layout proved hardest.
2026Extended the proactive-critique line with MMPCBench, benchmarking multimodal large language models on flawed inputs (not yet reviewed by Evolve).
2026OpenAlex records 19 works in 2026 — the profile's entire recent-three-year output — corresponding to an activity score of 87/100.

Papers in the Evolve catalog

Frequent co-authors

Yi Chang×3 Gengxu Li×2 Zhongru Chen×1 Jinzhe Li×1 Jinnan Li×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 indicators come from a single OpenAlex snapshot; citation-based dimensions (output, impact, focus, reproducibility) lack sufficient data, with h-index and i10-index recorded as 0.
The profile is name-resolved from public bibliographic data (identity: openalex); no ORCID is on file and 'Yuan Wu' is a common name, so record linkage is not guaranteed.
Affiliation metadata is inconsistent: the country field reads 'US' while the listed institution is Jilin University; affiliation has not been independently verified.
Evolve review coverage is partial: only two of the three catalog papers carry verdicts; MMPCBench is unreviewed and lacks a slug, keywords, and a reproducibility rating.