Peking University researcher in educational foundation models and RLVR data-policy evaluation for LLM post-training
2
papers here
2
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
1
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—
This axis measures total publication volume and productivity over time, but the scoring data is insufficient. Only countsByYear (7 works in 2025, 19 in 2026) and an h-index of 1 are on record, so output is not scored.
Impact—
This axis measures citation-based influence (h-index, i10-index, citation counts). The record shows h-index 1 and i10-index 0, insufficient for a reliable score given the very recent 2025-2026 publication window.
Focus0/100
The focus score sits at 0, reflecting a genuinely split portfolio: fieldDist assigns 0.5 to natural language processing / educational AI and 0.5 to machine learning (LLM post-training, RLVR). Topic data reinforces this breadth — Topic Modeling (19), NLP techniques (10), multimodal ML applications (10), plus smaller clusters in information retrieval (3) and robotics RL (3). The two catalogued 2026 papers sit in different fields, so current work spans two directions rather than one.
Collaboration46/100
Collaboration scores 46. The OpenAlex record shows 9 distinct coauthors in the recent window, with recurring partners including Hengyi Feng, Mingrui Chen, Hao Liang, Qihan Lin, Linzhuang Sun, Sizhe Qiu and Wentao Zhang. On both reviewed 2026 papers this researcher holds a middle author position without corresponding-author status, indicating contribution within a larger team rather than independent leadership. The moderate score reflects a small but stable co-author network.
Activity96/100
Activity scores 96, the strongest axis in the record. countsByYear shows 7 works in 2025 rising to 19 in 2026, and recent3yWorks totals 26 — a steep ramp in output within roughly two years. Both catalogued papers appeared as arXiv preprints in 2026. This pace is characteristic of an intensive early-stage research phase, though the volume is recent and the long-run track record is still short.
Reproducibility78/100
Reproducibility scores 78, computed from 2 reviewed papers (reproMean 78). DataFlex-RL rates high readiness: its 591-run matched-seed null study is designed for verification. OmniEdu rates medium — Evolve notes plausible capability-balanced SFT gains on Qwen 4-27B across K-12 benchmarks, but same-named train/eval sources and missing ablations temper the claims. The pattern suggests above-average attention to controlled, reportable experimental design.
Research summary
Meiyi Qiang is an early-stage researcher at Peking University whose recorded work splits evenly between natural language processing / educational AI and machine learning for LLM post-training (RLVR). The OpenAlex-linked record (ORCID 0009-0009-3009-2886) shows a steep activity ramp: 7 works in 2025, 19 in 2026, and 26 over the recent three-year window, built with 9 distinct coauthors. Topic counts cluster around Topic Modeling (19), NLP techniques (10) and multimodal machine learning applications (10), with smaller clusters in information retrieval and robotics RL. Evolve has reviewed two 2026 arXiv preprints. OmniEdu: Open Foundation Models for Learning and Teaching reports capability-balanced supervised fine-tuning that lifts Qwen 4-27B across K-12 benchmarks; Evolve judges the gains plausible but tempered by same-named train/eval sources and missing ablations, rating reproducibility readiness medium. DataFlex-RL: An Evaluation Platform for RLVR Data Policies presents a 591-run matched-seed null study concluding that RLVR data policies add no reproducible gain over uniform GRPO, scoped to one recipe and 300 steps; it rates high on reproducibility readiness. The citation footprint is still minimal — h-index 1, i10-index 0 — so influence is not yet scored. On both reviewed papers the researcher is a middle author without corresponding status, consistent with contributing inside an established team while an independent agenda takes shape.
Collaboration pattern
Collaboration is moderate (score 46): the OpenAlex window lists 9 distinct coauthors, with recurring partners Hengyi Feng, Mingrui Chen, Hao Liang, Qihan Lin, Linzhuang Sun, Sizhe Qiu and Wentao Zhang. On both reviewed 2026 papers the researcher appears as a middle author without corresponding status, indicating contribution within a team consistent with the listed Peking University affiliation rather than solo project leadership.
Research keywords
Highlights
Capability-balanced SFT lifts Qwen 4-27B across K-12 benchmarks
591-run matched-seed null result on RLVR data policies
A platform for auditing post-training data choices
Timeline
- 2025First recorded output burst: 7 works appear in the OpenAlex countsByYear record.
- 2026Output ramps to 19 works in a single year, bringing the recent three-year total to 26.
- 2026OmniEdu: Open Foundation Models for Learning and Teaching is posted to arXiv; Evolve reviews it with medium reproducibility readiness.
- 2026DataFlex-RL: An Evaluation Platform for RLVR Data Policies is posted to arXiv; Evolve rates its reproducibility readiness high.
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 data comes from OpenAlex only (identity: openalex); citation coverage for very recent 2026 arXiv work is likely incomplete.
- h-index 1 and i10-index 0 indicate a very early-stage record; output and impact axes lack sufficient data and are not scored.
- Only two papers are in the Evolve catalog, both 2026 arXiv preprints with middle authorship; the verdicts may not represent the full body of work.
- The reproducibility score (mean 78) rests on 2 reviewed papers; the profile is name-resolved from public bibliographic data via ORCID 0009-0009-3009-2886.