Peking University researcher on vision-language model reliability — hallucination mitigation and confidence calibration
2
papers here
1
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
0
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—
No aggregate output score can be computed yet: the record shows only 3 works indexed in 2026 and carries no venue-quality or full publication-list metadata. This axis would measure total publication volume and its distribution across years and venues.
Impact—
Impact cannot be scored: h-index and i10-index are both 0 and no citation counts are available for the indexed works. This axis would measure citation-based influence, such as per-paper citations and field-normalized benchmarks.
Focus100/100
The focus dimension scores 100, the maximum. The reviewed output converges tightly on multimodal machine learning: the field distribution assigns the sole classified paper to vision-language models, and SAVOR's keywords (multimodal LLMs, hallucination mitigation, confidence calibration, abstention) form one coherent agenda. Topic labels such as NLP and software engineering sit adjacent, but the 2026 arXiv record shows a sharply concentrated direction rather than scattered interests.
Collaboration53/100
Collaboration scores 53 on the 0-100 scale. The record lists 13 distinct coauthors, twelve of whom recur frequently — including Zhaolu Kang, Meixin Wu, and Yu Xue. Both reviewed 2026 arXiv papers place the researcher in a middle author position without corresponding-author status, consistent with contributor roles inside larger teams. The moderate score reflects a growing but not yet central network position; no solo-authored work appears in the catalog.
Activity40/100
Activity scores 40. OpenAlex records exactly 3 works in 2026 — the only year with any indexed output — and recent3yWorks is 3, meaning all recorded activity is concentrated in a single recent year. This pattern is typical of an early-career profile or a recently disambiguated identity; the score should move quickly as new submissions appear.
Reproducibility55/100
Reproducibility scores 55, computed as reproMean 55 over the single paper Evolve has assessed (reproPapers = 1). For SAVOR, reproducibility readiness is rated medium: the method reports ECE reductions across two backbones, but underspecified baseline-confidence elicitation and a statistically fragile group-ECE estimator mean reimplementation requires care. With only one assessed paper, the score is indicative rather than stable.
Research summary
Yingjie He is a researcher affiliated with Peking University whose public bibliographic record, resolved from OpenAlex, currently centers on multimodal machine learning — specifically vision-language models. The indexed profile is early-stage: OpenAlex lists 3 works in 2026, the only year with recorded output, and 3 works over the last three years; h-index and i10-index are both 0, so citation impact is not yet measurable. Topic assignments span Multimodal Machine Learning Applications (2 works), Advanced Neural Network Applications, Natural Language Processing Techniques, Software Engineering Research, and Teaching and Learning Programming, suggesting applied work reaching beyond a single subfield. Evolve has reviewed two 2026 arXiv papers. SAVOR reframes hallucination mitigation in multimodal LLMs as confidence calibration inside GRPO reinforcement learning; the paper reports striking ECE reductions across two backbones, but the review flags underspecified baseline-confidence elicitation, a statistically fragile group-ECE estimator, and a suspicious citation block, rating reproducibility readiness medium. The second, Modality Fault Lines, uses structural corruptions to expose fragile omni-modal reasoning and is not yet fully reviewed. Both papers list the researcher in a middle author position without corresponding status. With a focus score of 100, a collaboration score of 53 across 13 distinct coauthors, and a reproducibility score of 55 from one assessed paper, this is a coherent but young profile whose citation footprint and leadership roles have yet to accumulate.
Collaboration pattern
Both reviewed papers place Yingjie He in a middle author position without corresponding status, consistent with contributor roles in team-based AI research. The record lists 13 distinct coauthors, twelve of whom recur: Zhaolu Kang, Meixin Wu, Yu Xue, Qiming Shi, Lei Wei, Yidi Wang, Richeng Xuan, Zhichao Hu, Zixiu Ding, Zilin Zhao, Xinlang Kang, and Guansu Wang. No solo-authored work appears in the catalog; collaboration score is 53.
Research keywords
Highlights
SAVOR: confidence-calibrated reinforcement learning for multimodal hallucination mitigation
Probing fragile omni-modal reasoning with structural corruptions
A second 2026 arXiv paper, 'Modality Fault Lines: Structural Corruptions Reveal Fragile Omni-Modal Reasoning', uses structural corruptions to expose weaknesses in omni-modal reasoning. It is not yet fully reviewed in the Evolve catalog — no verdict, keywords, or reproducibility rating — and the researcher is again a middle author.
A sharply focused early portfolio on vision-language reliability
OpenAlex topic and field assignments concentrate on multimodal machine learning (vision-language models), yielding a focus score of 100. The profile shows 3 works in 2026, 3 works over the last three years, and 13 distinct coauthors — a coherent, early-stage research agenda within an active but developing network.
Timeline
- 2026OpenAlex records 3 works in 2026 — the only year with indexed output; h-index and i10-index are both 0.
- 2026SAVOR, on confidence-calibrated reinforcement learning for multimodal hallucination mitigation, appears on arXiv; Evolve reviews it and rates reproducibility readiness medium.
- 2026'Modality Fault Lines: Structural Corruptions Reveal Fragile Omni-Modal Reasoning' appears on arXiv, a second middle-author paper not yet fully reviewed.
- 2026recent3yWorks equals 3: all recorded activity in the last three years falls within 2026.
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 figures (3 works in 2026, 13 distinct coauthors, h-index 0) come from an OpenAlex-resolved identity; no ORCID is linked, so name disambiguation may be imperfect for this common name.
- Output and impact axes are marked missing: citation counts, venue tiers, and h-based metrics are not available in the input data.
- The reproducibility score of 55 is a mean over a single assessed paper (SAVOR); the second catalog paper has no review, keywords, or reproducibility rating yet.
- Author-position data covers only the two reviewed arXiv papers; venue acceptance and the complete publication list are unknown.