Early-stage Zhejiang University researcher on LLM post-training (agentic RL) 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: OpenAlex lists 10 recent works for this record, but the resolved publication history is too incomplete to compute an output score; this axis would measure productivity volume and its trend over time.
Impact—
Not scored: recorded h-index and i10-index are both 0, so there is no accrued citation signal to rank; this axis would measure citation-based influence once the 2026 works gather citations.
Focus0/100
Scores 0, the lowest concentration band: the OpenAlex field split is exactly 50/50 between machine translation evaluation and LLM post-training/agentic RL, and the topic list spreads across five clusters (topic modeling 9, NLP techniques 4, neural networks 3, logic and reasoning 3, stochastic gradient optimization 3). The two reviewed papers confirm the dual-track pattern; the record reads as exploratory rather than narrowly specialized.
Collaboration58/100
Scores 58, mid-range: the record shows 17 distinct coauthors with a stable recurring core of 12, including Weiming Lu, Yongliang Shen, Yiwen Qiu, and Daoxin Zhang. Both reviewed 2026 papers place Liu in a middle author position without corresponding status, consistent with contributing inside larger Zhejiang University teams rather than leading them.
Activity70/100
Scores 70, reflecting a concentrated recent burst: countsByYear contains only 2026, with 10 works in the last three years all falling in that single year, and both reviewed papers are 2026 arXiv preprints. The snapshot cannot show whether this pace predates 2026, so the score should be read as strong current activity rather than a long-term publishing habit.
Reproducibility38/100
Scores 38, the mean across the 2 papers Evolve reviewed. RetireOPD is rated medium: strong result tables (+12–19 points over GRPO on ALFWorld/WebShop) but single-run experiments with no variance reported. The translation-evaluation paper is rated low: CuRIL's decaying-hint RL is promising, yet it lacks significance tests and public data. Releasing code, data, and multi-seed results would raise this axis.
Research summary
Yizhou Liu is an early-stage researcher affiliated with Zhejiang University whose OpenAlex-resolved record sits at the intersection of two lines of work: natural language processing for machine translation evaluation and LLM post-training via agentic reinforcement learning — the profile's field split is exactly 50/50 between them. The snapshot records 10 works in the last three years, all dated 2026, alongside 17 distinct coauthors and recorded h-index and i10-index of 0, consistent with a career at its beginning. Topic signals include topic modeling (9), NLP techniques (4), neural networks (3), logic and reasoning (3), and stochastic gradient optimization (3). Evolve has reviewed two 2026 arXiv preprints. RetireOPD introduces self-retiring on-policy distillation for GRPO-trained agents, reporting +12–19 points over GRPO on ALFWorld and WebShop, though from single runs without variance. The second paper shows that metrics and LLM judges fail on social-media translation lacking cultural grounding, proposing CuRIL trained with decaying-hint reinforcement learning. In both papers Liu is a middle author rather than corresponding author, pointing to a contributor role within larger teams. Reproducibility readiness is mixed (mean 38 across the two): significance testing and public data are the recurring gaps. Overall this is an active, team-embedded profile whose independent impact is not yet measurable.
Collaboration pattern
Yizhou Liu's OpenAlex record lists 17 distinct coauthors, 12 of whom recur — including Weiming Lu, Yongliang Shen, Yiwen Qiu, Linjuan Wu, Dingming Li, Zixuan Wang, Haolei Xu, Ye Guo, Daoxin Zhang, Yan Yu, Zhengxi Lu, and Yichen Pan. In both reviewed 2026 papers Liu is a middle-position, non-corresponding author, indicating a contributor role within larger (apparently Zhejiang University) teams rather than an independent lead.
Research keywords
Highlights
Self-retiring on-policy distillation for GRPO-based LLM agents
Internalizing cultural reasoning for social-media translation evaluation
One toolkit, two domains: GRPO-based RL as the common thread
Both reviewed 2026 papers build on GRPO-style reinforcement learning — one for agentic training with on-policy distillation, one for internalizing cultural reasoning in a translation metric — making RL post-training the connective method of this profile.
Timeline
- 2026Reviewed arXiv preprint: RetireOPD — self-retiring on-policy distillation for agentic reinforcement learning (middle author).
- 2026Reviewed arXiv preprint: CuRIL — internalizing cultural reasoning for social-media translation evaluation (middle author).
- 2026OpenAlex snapshot records 10 works dated 2026 with 17 distinct coauthors — the profile's entire visible output year so far.
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 come from OpenAlex only; 'Yizhou Liu' is a common name, so this profile is name-resolved from public bibliographic data and may merge or split identities.
- Recorded h-index and i10-index are 0 and only 2026-dated works appear; output and impact axes are left unscored rather than estimated.
- Only 2 works carry Evolve review verdicts; conclusions about method quality and reproducibility rest on those two arXiv preprints alone.
- Source metadata lists the affiliation country as 'US' while the institution is Zhejiang University; the discrepancy is unresolved.