Early-career ML researcher at Oxford working on preference-alignment evaluation and real-device multimodal GUI agents.
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—
An output score is not yet available: the OpenAlex record is too early-stage (h-index 2, i10-index 0) to aggregate reliably. This axis would measure total works, first/corresponding-author share, and venue quality.
Impact—
An impact score is not yet available: citation signals are too sparse to benchmark (h-index 2, i10-index 0, no i10-qualified papers). This axis would measure citations, field-weighted influence, and uptake of the reviewed work.
Focus0/100
The focus axis scores 0, reflecting breadth rather than concentration: the OpenAlex field distribution splits evenly at 0.5 machine learning / LLM alignment and 0.5 multimodal GUI agents, and the topic list spreads across five clusters — adversarial robustness (6), machine learning and data classification (5), 3D shape modeling (3), domain adaptation and few-shot learning (3), and security and verification (3). This reads as an exploratory early-career spread rather than a settled specialization.
Collaboration78/100
Scoring 78, collaboration is this profile's strongest measured axis: the OpenAlex record lists 49 distinct coauthors over the recent three-year window (12 works), and 12 recurring partners are identified, including Lei Feng, Chris Russell, Tongliang Liu, Jindong Gu, and Suqin Yuan. Both cataloged 2026 papers are middle-authored within large teams, consistent with a highly collaborative contributor who has not yet appeared as corresponding author in the reviewed set.
Activity75/100
Scoring 75, activity is rising sharply: the OpenAlex counts-by-year series shows 2 works in each of 2023, 2024 and 2025, then 8 in 2026 — a four-fold jump that makes 2026 the most productive year on record. Of the 12 works in the recent three-year window, two are the arXiv preprints Evolve has reviewed, spanning LLM alignment and multimodal GUI agents, indicating an accelerating, current publishing tempo.
Reproducibility38/100
Scoring 38 (reproMean across 2 scored papers), reproducibility is the weakest measured axis. Evolve rates the preference-alignment study medium readiness, while BlueLM-GUI is rated low: its headline 87.4 comes from the authors' own benchmark and one ablation is confounded, with real-device training hard to replicate. Releasing code, prompts and evaluation harnesses would most directly lift this axis.
Research summary
Runqi Lin (ORCID 0009-0000-6607-7754) is an early-career researcher at the University of Oxford whose OpenAlex record (h-index 2, i10-index 0) shows a steep recent ramp: 2 works in each of 2023, 2024 and 2025, rising to 8 in 2026, with 12 works across the recent three-year window. The field distribution splits evenly at 0.5 machine learning / LLM alignment and 0.5 artificial intelligence — multimodal GUI agents, while topics span adversarial robustness (6), machine learning and data classification (5), 3D shape modeling and analysis (3), domain adaptation and few-shot learning (3), and security and verification in computing (3). Two 2026 arXiv preprints anchor the reviewed record. A middle-authored alignment study argues that preference-aligned models (RLHF/DPO) drift away from human response distributions — a Turing-test gap — unless a restrictive reward condition holds; Evolve credits solid 7-8B evidence while cautioning that frontier claims outrun the scope. The BlueLM-GUI technical report, also middle-authored, couples real-phone training with an agentic reinforcement learning flywheel (GRPO), reporting 87.4 on its own benchmark and 84.9 on AndroidWorld. Radar dimensions score collaboration 78 and activity 75, focus 0 (broad spread), and reproducibility 38 (reproMean across 2 papers); output and impact are not yet scorable.
Collaboration pattern
Runqi Lin's record shows 49 distinct coauthors within the recent three-year window and a stable group of 12 recurring partners — among them Suqin Yuan, Muyang Li, Guanzhe Hong, Jindong Gu, Lei Feng, Chris Russell, Tongliang Liu, Tong Ye, Kunyang Han, Guozhi Wang, Longqiang Luo and Zhifeng Ding. Both reviewed 2026 papers are middle-authored in large teams, and the collaboration axis scores 78, the profile's highest; no corresponding-author role appears in the reviewed set.
Research keywords
Highlights
Evidence that preference alignment can pull models away from human behavior
Real-device flywheel for self-improving mobile GUI agents
Breadth across trustworthy-ML topics
OpenAlex topic counts rank adversarial robustness first (6 mentions), followed by machine learning and data classification (5), with 3D shape modeling, domain adaptation and few-shot learning, and security and verification at 3 each — a spread that pairs with the even 0.5/0.5 field split.
Timeline
- 2023OpenAlex record begins: 2 indexed works.
- 20242 further indexed works, holding a steady early tempo.
- 20252 indexed works; the recent three-year window now spans 12 works.
- 2026Most productive year on record: 8 indexed works, including both reviewed arXiv preprints — the preference-alignment Turing-test study and the BlueLM-GUI technical report.
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 a single OpenAlex record (identity: openalex); h-index 2 and i10-index 0 reflect an early-career profile, not a mature citation trail.
- Output and impact dimension scores are unavailable in the source data and are reported as missing rather than estimated.
- Evolve has reviewed only 2 of this researcher's papers, so topic and field distributions may not capture the full oeuvre.
- Affiliation is listed as University of Oxford with country AU as supplied by the source; the profile is name-resolved from public bibliographic data (ORCID 0009-0000-6607-7754).