Cheng Qian

University of Illinois Urbana-Champaign · United States

Unclaimed · AI-structured profileOpenAlex

Early-career UIUC researcher in LLM post-training and knowledge updating, with a steep 2026 publication ramp

papers here

reviewed by Evolve

first-author

corresponding

—

works · OpenAlex

h-index · OpenAlex

Bibliometric figures come from OpenAlex (2026-10-17) 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—

Output scoring is unavailable: this profile lacks the verbatim publication-count and venue-quality data needed to compute it, so the axis — which would measure total works, venue tiers, and first/corresponding-author share — is left unscored rather than estimated.

Impact—

Impact scoring is not yet computable: citation-linked evidence is essentially absent (h-index 1, i10-index 0), so the axis — meant to capture citations, citation velocity, and downstream uptake of the work — is reported as missing rather than scored.

Focus0/100

The focus score is 0, reflecting an OpenAlex topic spread rather than a single line: topic modeling (10 works), AI-based problem solving and planning (8), reinforcement learning in robotics (5), domain adaptation and few-shot learning (5), and NLP techniques (3). Evolve's own field attribution is narrower, splitting 0.5/0.5 between LLM post-training and LLM knowledge updating, so topical breadth looks larger in the topic data than in the curated field view.

Collaboration66/100

Collaboration scores 66 on a 0-100 scale: the OpenAlex record lists 27 distinct coauthors over the recent three-year window, and the snapshot highlights twelve recurring partners, including Yumeng Wang, Yuchen Wu, Zhiyuan Fan, Hyeonjeong Ha, and Heng Ji. All three catalog papers place this researcher in a middle author position, consistent with contributing within larger teams rather than solo-led projects.

Activity91/100

Activity scores 91, the strongest axis in this profile. The counts-by-year series shows one work in 2024 and one in 2025, jumping to 20 works in 2026, for 22 works across the recent three-year window — an unusually steep ramp concentrated in a single year. The three 2026 arXiv papers in the Evolve catalog are consistent with this burst of output.

Reproducibility78/100

Reproducibility scores 78, computed from two papers with reproducibility assessments (reproMean 78, reproPapers 2). In the catalog, the on-policy distillation paper is rated high reproducibility readiness — solid at 1.5B-7B scale but single runs only — while the knowledge-updating paper is rated medium, with rigorous controls yet noisy source-side evidence and flip rates near ceiling.

Research summary

Cheng Qian is a researcher at the University of Illinois Urbana-Champaign whose profile is name-resolved from public OpenAlex bibliographic data. The snapshot shows a steep publication ramp: 22 works in the recent three-year window, 20 of them in 2026, against one each in 2024 and 2025, while citation indicators remain at very early-stage levels (h-index 1, i10-index 0). OpenAlex field attribution splits evenly between machine learning — LLM post-training and natural language processing (LLM knowledge updating); assigned topics span topic modeling (10), AI-based problem solving and planning (8), reinforcement learning in robotics (5), domain adaptation and few-shot learning (5), and NLP techniques (3). The Evolve catalog lists three 2026 arXiv papers, each with this researcher as a middle author. 'Rethinking On-Policy Distillation of Large Language Models II: One Training Example' finds that a single training query recovers most of on-policy distillation's gain and carries a high reproducibility-readiness rating. 'Popular Knowledge Propagates More Errors in LLM Knowledge Updating' reports that popular facts suffer the greatest collateral damage during updates, rated medium. 'Evaluating the Hidden Costs of Personalization in Large Language Models' is cataloged but not yet reviewed. The collaboration footprint is broad: 27 distinct coauthors in three years, including recurring partners such as Yumeng Wang, Zhiyuan Fan, and Heng Ji.

Collaboration pattern

Collaboration is broad and team-based: 27 distinct coauthors appear in the OpenAlex record over the recent three-year window, with twelve recurring partners including Yumeng Wang, Yuchen Wu, Zhiyuan Fan, Hyeonjeong Ha, Shujin Wu, and Heng Ji. All three Evolve catalog papers list this researcher as a middle author rather than first or corresponding, consistent with contributing within larger research teams.

Research keywords

LLM post-trainingon-policy distillationknowledge distillationLLM knowledge updatingknowledge editingcatastrophic forgettingdata efficiencyLoRA fine-tuningLLM personalizationreinforcement learning in robotics

Highlights

One training example recovers most of on-policy distillation's gain
Evolve's review finds that at 1.5B-7B scale a single training query captures most of on-policy distillation's benefit, attributed to state coverage and slow absorption; the paper's reproducibility readiness is rated high, with the caveat of single runs only.
Popular facts take the biggest hit in LLM knowledge updating
The study reports that widely known facts suffer the most collateral damage when LLM knowledge is updated; Evolve credits rigorous controls but notes noisy source-side evidence and flip rates near the ceiling. Reproducibility readiness: medium.
Early catalog entry on the hidden costs of LLM personalization
The 2026 arXiv paper 'Evaluating the Hidden Costs of Personalization in Large Language Models' lists this researcher as a middle author; Evolve has not yet published a review verdict, keywords, or reproducibility rating for it.

Timeline

2026Co-authored 'Rethinking On-Policy Distillation of Large Language Models II: One Training Example', showing one training query recovers most of on-policy distillation's gain.
2026Co-authored 'Popular Knowledge Propagates More Errors in LLM Knowledge Updating', finding that popular facts suffer the most collateral damage during updates.
2026Co-authored 'Evaluating the Hidden Costs of Personalization in Large Language Models', listed in the Evolve catalog without a review verdict yet.

Papers in the Evolve catalog

Frequent co-authors

Yumeng Wang×1 Yuchen Wu×1 Zhiyuan Fan×1 Hyeonjeong Ha×1 Shujin Wu×1 Jiayu Liu×1 Heng Ji×1 Ge Wang×1 Zixuan Fu×1 Bingxiang He×1 Yuxin Zuo×1 Haohuan Huang×1

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.

All bibliometrics come from a single OpenAlex record (identity: openalex); no ORCID is linked, so this profile is name-resolved from public bibliographic data, and Cheng Qian is a common name.
Output and impact dimensions are unscored: citation coverage is minimal (h-index 1, i10-index 0), typical of a very early-stage record.
Only two papers carry Evolve reproducibility assessments; the third catalog paper has no verdict, keywords, or slug yet.
Counts by year (1 in 2024, 1 in 2025, 20 in 2026) reflect OpenAlex coverage and may lag or over-count arXiv preprints.