Songlin Yang
Hong Kong University of Science and Technology · Hong Kong SAR China
AI-generated media evaluation researcher at HKUST: image quality assessment and rubric-guided video reward modeling.
2
papers here
2
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
0
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-09) 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—
Aggregate output is not yet scorable: the OpenAlex record shows an h-index of 0 and only 4 works in the past three years, all dated 2026. This axis would measure total publication volume and its distribution over time.
Impact—
Impact data is insufficient: the snapshot reports an h-index of 0 and an i10-index of 0, and both catalog papers are 2026 arXiv preprints too new for citation accumulation. This axis would measure citations and uptake of the methods.
Focus0/100
Focus scores 0, the lowest reading, and the raw field distribution explains why: work is split exactly 50/50 between computer-vision image quality assessment and machine-learning video generation evaluation. OpenAlex topics span five clusters (image retrieval and classification, image and video quality assessment, multimodal machine learning, digital media forensics, neural networks, at 2/2/2/2/1 works). The split is real, but both halves share one theme: judging generated media quality.
Collaboration51/100
Collaboration scores 51, a mid-range reading. The OpenAlex record lists 12 distinct coauthors, with recurring partners including Bo Peng, Jing Dong, Zhenchen Tang, Zichuan Wang, Leilei Cao and Fengjie Zhu. On both reviewed 2026 papers Yang sits in a middle author position without corresponding-author role, consistent with a contributor pattern inside small teams rather than a solo or lead-first one.
Activity47/100
Activity scores 47. The OpenAlex record shows 4 works in the past three years, and the counts-by-year series attributes all 4 to 2026, with both catalog papers posted as arXiv preprints that year. This is a compact but recent burst concentrated on evaluation of AI-generated media; the short series limits what the score can say about sustained output.
Reproducibility38/100
Reproducibility scores 38, computed as the mean over 2 reviewed papers. The agentic IQA paper rates medium — its protocol is training-free but the central evaluation runs on a self-built 180-image set without statistical tests — while RewardVerse rates low, with headline tables missing despite convincing ablations. Neither barrier is fatal, but independent replication currently requires materials the papers do not yet release.
Research summary
Songlin Yang is affiliated with Hong Kong University of Science and Technology and works on evaluating AI-generated media, a profile built from an OpenAlex bibliometric snapshot plus two Evolve-reviewed 2026 arXiv preprints. The record shows 4 works in the past three years, all attributed to 2026, and 12 distinct coauthors; h-index and i10-index currently read 0, so no impact scoring is possible yet. The research splits evenly — 50/50 in the field distribution — between computer-vision image quality assessment and machine-learning video generation evaluation, with OpenAlex topics also touching image retrieval and classification, multimodal machine learning, and digital media forensics. RewardVerse (2026, middle authorship) proposes rubric-guided policy optimization for video reward modeling, targeting scalar drift with GRPO-style training and soft-logits readout; Evolve finds the ablations convincing but headline tables missing, rating reproducibility low. The evolutionary agentic IQA paper (2026, middle authorship) uses multi-agent workflows to reduce holistic bias in open-ended image quality perception, reporting HOR falling from 44.4% to 8.6%, though the central evaluation rests on a self-built 180-image set without statistical tests (reproducibility medium). The blended reproducibility score of 38 across the two reviewed papers marks an early-stage record: technically current, focused on a fast-moving evaluation niche, but awaiting citation accumulation and stronger release of artifacts.
Collaboration pattern
The OpenAlex record lists 12 distinct coauthors. Recurrent partners include Bo Peng, Jing Dong, Zhenchen Tang, Zichuan Wang, Leilei Cao and Fengjie Zhu, with other names such as Yang Li, Xiaotong Zhao, Shuai Li, Haotian Fan and Alan Zhao. Both reviewed 2026 papers list Yang in a middle author position and not as corresponding author, suggesting a contributing role within multi-author teams working on media-quality evaluation rather than a solo or consistently lead-author pattern.
Research keywords
Highlights
Rubric-guided reward modeling for video generation
Evolutionary multi-agent pipeline for image quality perception
Timeline
- 2026OpenAlex attributes all 4 recorded works of the past three years to 2026, marking the visible start of this publication record.
- 2026RewardVerse preprint on rubric-guided policy optimization for video reward modeling appears on arXiv, with Yang in a middle author position.
- 2026Evolutionary agentic approach for open-ended image quality perception posted to arXiv; Evolve review records holistic-bias rate falling from 44.4% to 8.6% on a self-built 180-image set.
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 indicators come from an OpenAlex snapshot only: h-index and i10-index read 0 and no citation-impact series is available.
- Only the 2 arXiv 2026 preprints reviewed by Evolve are individually assessed; other works behind the count of 4 recent works are not evaluated here.
- No ORCID is linked and 'Songlin Yang' is a common name, so identity is resolved from public bibliographic data and records may merge or split.
- Reproducibility ratings cover only the 2 reviewed papers (mean 38), and no peer-review venue information is available for either preprint.