Zhejiang University researcher in agentic RL for LLM post-training and machine translation evaluation
2
papers here
2
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
0
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-16) 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 scored: this identity record lacks reliable cumulative output totals, so publication volume, venue mix, and productivity trends cannot be computed yet. This axis would measure total scholarly output and its distribution across venues and years.
Impact—
Not scored: citation counts, h-index, and i10-index are unavailable or recorded as 0 in this snapshot, so influence cannot be assessed. This axis would measure citation-based impact such as h-index, i10-index, and citation trajectory.
Focus0/100
The focus score is 0, and the raw field distribution explains why: work splits evenly (0.5/0.5) between machine translation evaluation and LLM post-training/agentic RL. OpenAlex topics are similarly spread — topic modeling, NLP techniques, and multimodal ML (7 each), plus RL in robotics (5) and neural network applications (4). This is a two-front, spread-topic profile rather than deep single-area specialization.
Collaboration58/100
Collaboration scores 58. The OpenAlex record lists 17 distinct co-authors, with a dense recurring core of twelve frequent collaborators including Yizhou Liu, Weiming Lu, Yiwen Qiu, and Linjuan Wu. On both 2026 cataloged papers Shen sits in the last author position without a corresponding-author flag — a placement commonly associated with supervisory roles, though the record does not confirm that status.
Activity91/100
Activity is the strongest axis at 91. OpenAlex records 22 works in the recent three-year window, and the only year-level count provided is 2026 with 22 works, indicating a concentrated, high-tempo recent phase. Both cataloged papers are 2026 arXiv releases, consistent with this picture of intense current output; earlier-year counts are absent from the snapshot, so the longer-term trend cannot be read.
Reproducibility38/100
Reproducibility scores 38, averaged over the 2 cataloged papers Evolve assessed. RetireOPD is rated medium: strong result tables (+12–19 points over GRPO on ALFWorld/WebShop) but single-run experiments with no reported variance. The CuRIL translation-evaluation paper is rated low: promising decaying-hint RL, but no significance tests and no public data. Strengthening statistical reporting and artifact release would most directly lift this axis.
Research summary
Yongliang Shen is a researcher at Zhejiang University whose public bibliographic record (OpenAlex) shows a profile concentrated in two areas of roughly equal weight: natural language processing centered on machine translation evaluation, and machine learning focused on LLM post-training and agentic reinforcement learning (field distribution 0.5/0.5). Evolve's catalog contains two 2026 arXiv papers on which Shen appears in the last author position without corresponding-author designation. The first, RetireOPD (2026), introduces self-retiring on-policy distillation for agentic RL, reporting +12–19 point gains over GRPO on ALFWorld/WebShop, though from single runs without variance. The second, the CuRIL work (2026), shows that standard metrics and LLM judges can reward the worst translations on social-media machine translation, internalizing cultural reasoning through decaying-hint reinforcement learning. Activity signals are strong: 22 works in the recent three-year window (the only year-level count provided is 2026 with 22), an activity score of 91, 17 distinct co-authors, and a collaboration score of 58. Reproducibility is the weakest axis (38, averaged over 2 assessed papers), rated medium and low. Citation indices are recorded as 0 in this snapshot, most likely reflecting an incomplete identity record rather than a reliable impact measure.
Collaboration pattern
The collaboration pattern centers on a stable Zhejiang University group: 17 distinct co-authors in the OpenAlex record, with twelve frequent collaborators — Yizhou Liu, Weiming Lu, Yiwen Qiu, Linjuan Wu, Dingming Li, Zixuan Wang, Haolei Xu, Ye Guo, Daoxin Zhang, Yan Yu, Zhengxi Lu, and Yichen Pan. On both 2026 papers Shen takes the last author slot without a corresponding-author flag, a position commonly tied to supervisory roles.
Research keywords
Highlights
Self-retiring on-policy distillation lifts agentic RL agents
Exposing metric failure in social-media translation evaluation
A concentrated, dual-front 2026 output burst
The OpenAlex snapshot places 22 recent three-year works in 2026, split evenly between machine translation evaluation and LLM post-training/agentic RL, with 17 distinct co-authors — a high-tempo year represented in the catalog by both reviewed papers.
Timeline
- 2026OpenAlex attributes all 22 recorded recent three-year works to 2026, the only year with a count in the snapshot.
- 2026Releases RetireOPD on arXiv, proposing self-retiring on-policy distillation for agentic RL with +12–19 point gains over GRPO on ALFWorld/WebShop.
- 2026Releases the CuRIL social-media machine translation evaluation work, showing metrics and LLM judges reward the worst translations without cultural grounding.
- 2026Evolve assesses reproducibility across both papers (mean 38; readiness medium and low), citing single-run results and missing significance tests.
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.
- Profile is name-resolved from public bibliographic data (OpenAlex, identity: openalex); no ORCID was available for cross-checking, and the name may be shared by multiple researchers.
- h-index and i10-index are recorded as 0 and the output and impact axes are unscored; this most likely reflects an incomplete or unmerged identity record rather than measured impact.
- The snapshot lists country 'IL' alongside Zhejiang University, an affiliation–country mismatch that could not be resolved from the provided data.
- Evidence rests on two 2026 arXiv papers in Evolve's catalog; no peer-reviewed venue, citation, funding, or career information was available in the input data.