Peking University researcher bridging educational AI and LLM post-training (RLVR), with a steep recent output ramp
2
papers here
2
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
2
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 yet scored — the platform flags insufficient data on publication volume. This axis would measure total output, yearly cadence, and productivity trend; only fragmentary counts (9 works in 2026; h-index 2) are on record.
Impact—
Not yet scored — citation influence data is too thin for a reliable estimate (h-index 2, i10-index 2 on record). This axis would measure citation counts, citing works, and downstream uptake of the two 2026 releases.
Focus0/100
Scores 0 on the focus axis in this snapshot, reflecting an unusually even field split: fieldDist assigns 0.5 to natural language processing / educational AI and 0.5 to machine learning (LLM post-training, RLVR). Topic counts reinforce the spread — topic modeling (11), multimodal ML applications (6), image and video retrieval (3), NLP (2), domain adaptation / few-shot learning (2) — so the record reads as dual-track rather than narrowly specialized.
Collaboration46/100
Scores 46 on the collaboration axis. The OpenAlex snapshot lists 9 distinct co-authors, with Wentao Zhang and Hao Liang recurring most often in the frequency list, alongside Meiyi Qiang, Mingrui Chen, Qihan Lin, Linzhuang Sun, and Sizhe Qiu. Both Evolve-reviewed 2026 papers place Feng in a middle authorship position without corresponding-author role, indicating a stable team contributor rather than an independent lead so far.
Activity72/100
Scores 72 on the activity axis. countsByYear shows 2 works in 2025 and 9 in 2026, and recent3yWorks totals 11 — a steep, current ramp concentrated in the last two years. The 2026 cluster includes both Evolve-reviewed releases (OmniEdu and DataFlex-RL, both arXiv), placing the profile in an active, accelerating phase rather than a plateau.
Reproducibility78/100
Scores 78, computed from 2 reproducibility-assessed papers (reproMean 78). DataFlex-RL is rated high: a 591-run matched-seed null study of RLVR data policies is transparent about scope (one recipe, 300 steps). OmniEdu is rated medium: capability-balanced SFT gains on Qwen 4-27B are plausible, but same-named train/eval sources and missing ablations temper them.
Research summary
Hengyi Feng is a researcher at Peking University whose bibliometric record, resolved from OpenAlex, centers on two adjoining areas: natural language processing for educational AI, and machine learning for LLM post-training with reinforcement learning from verifiable rewards (RLVR). The snapshot splits the field distribution evenly (0.5/0.5) between these tracks. Topic frequencies are led by topic modeling (11 works), followed by multimodal machine learning applications (6), image and video retrieval (3), NLP techniques (2), and domain adaptation / few-shot learning (2). Recent activity is high: 11 works in the last three years, including 9 recorded in 2026 and 2 in 2025. Citation-linked metrics remain early-stage — h-index 2 and i10-index 2 — so output and impact are not scored. Two Evolve-reviewed 2026 papers define the current agenda: DataFlex-RL, a 591-run matched-seed evaluation platform reporting that RLVR data policies add no reproducible gain over uniform GRPO (high reproducibility readiness), and OmniEdu, capability-balanced SFT lifting Qwen 4-27B across K-12 benchmarks (medium readiness). In both, Feng is a middle-position, non-corresponding author, consistent with a team contributor within a nine-distinct-coauthor network.
Collaboration pattern
The OpenAlex snapshot lists 9 distinct co-authors over the recent window. Wentao Zhang and Hao Liang each appear twice in the frequent-coauthor list — the strongest repeat ties — alongside Meiyi Qiang, Mingrui Chen, Qihan Lin, Linzhuang Sun, and Sizhe Qiu. Both Evolve-reviewed 2026 papers carry a middle, non-corresponding authorship position, and the collaboration dimension scores 46: Feng currently operates as a contributing team member within a compact collaboration network rather than as an independent lead.
Research keywords
Highlights
DataFlex-RL: a 591-run null result on RLVR data policies
OmniEdu: capability-balanced SFT for K-12 foundation models
Timeline
- 2001OpenAlex counts 2 works under this name in 2001 — sharply inconsistent with the 2025–2026 burst and most plausibly a name-disambiguation artifact; treat with caution.
- 2025countsByYear records 2 works in 2025, marking the onset of the current publication ramp.
- 20269 works recorded in 2026, including both Evolve-reviewed arXiv releases — OmniEdu and DataFlex-RL — in middle authorship positions with no corresponding role.
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; no ORCID is linked, so this profile is name-resolved from public bibliographic data and common-name collisions are possible.
- The 2001 entry in countsByYear (2 works) is inconsistent with the otherwise 2025–2026 activity burst and is most plausibly a disambiguation artifact.
- Output and impact dimensions are unscored (dimMissing): the recorded h-index 2 and i10-index 2 are too thin to support a reliable estimate.
- Reproducibility is computed from only 2 assessed papers (reproMean 78), and Evolve has reviewed 2 catalog papers, so the evidence base is small.