Sichuan University researcher advancing training-free VLM inference and test-time adaptation for robust multimodal AI.
3
papers here
3
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
13
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-15) 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 score is available yet: the platform lacks complete venue, type and volume records. This axis would measure publication productivity — indexed work counts, yearly consistency, and venue quality — so it is left unscored rather than estimated.
Impact—
Impact is not scored: citation context is insufficient for a platform rating. This axis would measure citation influence through field-normalized comparisons and uptake over time; the recorded h-index of 13 is not converted into a score here.
Focus0/100
The focus score is 0/100, reflecting an OpenAlex field distribution split almost exactly three ways — efficient VLM inference, test-time adaptation and robustness, and vision-language modeling, each about 33.3%. Topic counts add breadth: advanced neural network applications (24), domain adaptation and few-shot learning (18), and medical image segmentation techniques (15). The portfolio reads as spread rather than concentrated, though the three current fields are thematically adjacent.
Collaboration51/100
Collaboration scores 51/100 on a modest base: 12 distinct coauthors on record, with a recurring core of Zhenbin Wang, Wei Huang, Yan Wang, Lei Zhang, Zhenwei Zhang, and Zhao Zhang. All three 2026 catalog papers list Wang in a middle author position with no corresponding-author record, suggesting steady participation within small, stable teams rather than leadership of a large external network.
Activity89/100
Activity is the strongest axis at 89/100. OpenAlex counts rise from 1-2 works per year in 2017-2020 to 6 in 2021, 7 in 2024 and 10 in 2026, with 20 works in the last three years. The 2026 catalog additions — three arXiv papers on VLM inference and test-time adaptation — underscore a current, fast-moving output rhythm.
Reproducibility70/100
Reproducibility averages 70/100 across the three Evolve-reviewed papers (reproPapers: 3). StackTok is rated high readiness — training-free, principled and well specified — while MASA and AREA are medium: gains are consistent but modest, and neither reports statistical testing or latency figures, which tempers but does not negate a solid overall readiness.
Research summary
Lituan Wang is a researcher at Sichuan University whose work spans computer vision, multimodal machine learning, and machine-learning robustness, with an OpenAlex-recorded profile showing an h-index of 13 and an i10-index of 14. Indexed output spans 2017-2026 and has accelerated markedly: from one or two works per year in 2017-2020 to 6 in 2021, 7 in 2024 and 10 in 2026, with 20 works in the most recent three years — a high activity score of 89/100. The topic record is broad, led by advanced neural network applications (24 works), domain adaptation and few-shot learning (18), and medical image segmentation techniques (15), while the current field distribution splits almost evenly across efficient VLM inference, test-time adaptation and robustness, and vision-language modeling. Three 2026 arXiv papers reviewed by Evolve — StackTok, MASA and AREA — concentrate on training-free, inference-time methods for vision-language models and test-time adaptation; Evolve rates their reproducibility at 70/100 on average, with StackTok judged high-readiness. Collaboration is mid-range at 51/100, with 12 distinct coauthors and a small recurring core. Output and impact scores are not yet computable from the available data.
Collaboration pattern
The collaboration axis scores 51/100, consistent with a small, stable network: 12 distinct coauthors on record. A recurring core — Zhenbin Wang, Wei Huang, Yan Wang, Lei Zhang, Zhenwei Zhang, and Zhao Zhang — appears repeatedly in the frequent-coauthor list. On the three 2026 catalog papers Wang sits in a middle author position with no corresponding-author record, suggesting contribution within team-led projects rather than independent group leadership.
Research keywords
Highlights
StackTok: budget-adaptive visual token selection tops training-free baselines
MASA breaks the self-referential loop in wild test-time adaptation
AREA: adaptive evidence selection wins across all nine KB-VQA settings
Timeline
- 2017Earliest indexed works appear: 2 publications recorded in OpenAlex for 2017.
- 2021Output steps up to 6 works, the first clear acceleration after years of 1-2 per year.
- 20247 works recorded, sustaining the elevated output pace of recent years.
- 2026Most prolific year yet: 10 works recorded, including three arXiv papers reviewed by Evolve — StackTok, MASA and AREA — on training-free VLM inference and test-time adaptation.
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 are sourced from OpenAlex only; citation and venue records are not independently verified.
- Output and impact radar axes are unscored because platform data is insufficient; the recorded h-index (13) and i10-index (14) are raw counts, not normalized impact measures.
- Evolve's review covers three 2026 arXiv papers with middle authorship; earlier works (2017-2025) appear only in aggregate topic and year counts, not as individually reviewed papers.
- Identity is name-resolved from public OpenAlex records; the frequent-coauthor list contains duplicates and the author's own name, so collaboration figures should be read as approximate.