CAS-affiliated researcher building verifiable LLM-agent environments and learner-outcome benchmarks for AI education
3
papers here
2
reviewed by Evolve
0
first-author
0
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—
No output score is available yet: total publication counts, venue mix, and year-by-year output trajectory are absent from the current data. This axis would measure publication volume and consistency over time once bibliometric coverage is sufficient.
Impact—
Impact cannot be scored: no citation counts, h-index, or venue-level uptake metrics are provided. This axis would measure how widely the two benchmarks are cited and adopted by others, and awaits citation coverage before any estimate is made.
Focus0/100
Focus scores 0 on the 0-100 scale, a direct readout of the raw field distribution: coded activity splits evenly, 50% machine learning / LLM agents and 50% natural language processing (AI in education). This is a dual-track profile rather than a diffuse one — two benchmark-building lines, plus an option-critic analysis sitting outside both coded fields.
Collaboration74/100
Collaboration scores 74, grounded in 41 distinct coauthors recorded in the snapshot — a wide network relative to a three-paper catalog. The frequent-coauthor list (Shuai Bai, Jiayong Deng, Sicheng Fan, Chang Gao, among others, with Mianqiu Huang listed twice) suggests recurring team-based work on shared arXiv projects. The axis captures co-authorship breadth, not seniority or role distribution within teams.
Activity—
Activity is unscorable: recent-year publication frequency and cadence are not present in the data. This axis would measure ongoing research tempo — papers per year and recency of output — and awaits fuller bibliometric coverage before a score can be computed.
Reproducibility78/100
Reproducibility scores 78, computed as the mean readiness over exactly two reviewed papers: RecreationWorld rates high (five-platform, verifiable-reward environments built for reuse) and SLATE rates medium. The signal favors open, benchmark-style artifacts, but with only two scored papers the average is a thin base — one more paper could move it materially.
Research summary
Yizhen Jiang is affiliated with the Chinese Academy of Sciences and works across two equally weighted tracks: machine learning / LLM agents and natural language processing for education, each carrying 50% of the coded field distribution. The catalog centers on benchmark construction. RecreationWorld (2026, arXiv) contributes scalable, verifiable recreation-task environments for hybrid computer-use agents together with an SFT training pipeline; Evolve judges the evaluation design rigorous while noting that transfer gains lack statistical controls, and rates its reproducibility readiness high. SLATE (2026, arXiv) asks whether AI-generated slides are educationally effective, claiming the first learner-outcome benchmark for teaching-slide quality and knowledge acquisition; the design is rigorous, but the headline D1/D2 dissociation rests on only ten systems, and reproducibility readiness is medium. A third 2026 paper, an option-critic analysis written from the last-author position, has not yet been reviewed by Evolve. Radar scores stand at 74 for collaboration (41 distinct coauthors) and 78 for reproducibility (mean over two papers); focus registers 0, a direct readout of the 50/50 field split, while output, impact, and activity lack sufficient data. The profile is name-resolved from public bibliographic data; no ORCID is attached.
Collaboration pattern
The collaboration score of 74 rests on 41 distinct coauthors in the snapshot — broad for a three-paper catalog. Recurring names include Shuai Bai, Jiayong Deng, Sicheng Fan, Yikun Fu, Chang Gao, Mianqiu Huang, Yuheng Jing, Dehui Kong, Keliang Li, Ning Li, and Wanli Li, pointing to stable team-based arXiv projects rather than one-off pairings. Roles in the reviewed set are middle or last author; no first-author or corresponding-author entry appears.
Research keywords
Highlights
RecreationWorld: five-platform verifiable environments for hybrid computer-use agents
SLATE: first learner-outcome benchmark for AI-generated teaching slides
Option-critic analysis from the last-author position
The 2026 paper 'When Do Options Help? Policy Necrosis and Redundant Coverage in Option-Critic' lists the researcher as last author. It has not yet received an Evolve review, so no verdict, keywords, or reproducibility rating are on record.
Timeline
- 2026Publishes RecreationWorld on arXiv: scalable, verifiable recreation-task environments and an SFT pipeline for hybrid GUI/coding agents.
- 2026Publishes SLATE on arXiv: a benchmark testing whether AI-generated slides produce real learner knowledge gains in language teaching.
- 2026Appears as last author on the arXiv option-critic study 'When Do Options Help? Policy Necrosis and Redundant Coverage in Option-Critic'.
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.
- Identity is name-resolved from public bibliographic data; no ORCID is attached and 'Yizhen Jiang' is a common name, so affiliation-based matching may miss or merge records.
- All bibliometric scores derive from a single OpenAlex-derived source; the output, impact, and activity axes remain unscored due to insufficient coverage.
- The reproducibility score of 78 averages only two reviewed papers, making it a thin-base estimate rather than a stable rating.
- The evidence base is just three 2026 arXiv papers, one of which (the option-critic study) has no Evolve verdict; no country or venue information is provided.