Peking University researcher in LLM-based multi-agent systems: RL-shaped agent topologies and memory-agent co-evolution
2
papers here
2
reviewed by Evolve
2
first-author
0
corresponding
—
works · OpenAlex
0
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-15) 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—
Axis unscored: publication-volume data is insufficient. The record holds only 4 works, all from 2026, too short to measure cumulative output, venue quality, and authorship roles fairly. Evolve withholds a score rather than extrapolating from a single year.
Impact—
Axis unscored: citation data is effectively empty — h-index and i10-index are both recorded as 0 with no citation history. This axis would measure citations, h-index, and downstream uptake once the 2026 preprints accumulate readership; no estimate is offered here.
Focus100/100
Focus scores a maximal 100, and the underlying data supports it: the field distribution lists a single field — machine learning (LLM-based multi-agent systems) — and topic counts cluster tightly, with AI-based Problem Solving and Planning at 3 and Multi-Agent Systems and Negotiation at 2, versus single occurrences of topic modeling, semantic web, and robotics RL. This is a deliberately narrow agenda around LLM multi-agent design.
Collaboration51/100
The collaboration dimension scores 51. The raw record shows 12 distinct coauthors across the cataloged work, with a stable recurring core — Xunkai Li, Rong-Hua Li, Ziheng Yi, Zekai Chen, Minghao An, Zhanke Liu, and Guangyuan He — suggesting a small standing team rather than wide ad-hoc networking; the moderate score reflects a compact but consistent partner set behind first-author preprints.
Activity47/100
Activity scores 47. The observable window is a single year: countsByYear lists only 2026 with 4 works, and recent3yWorks is likewise 4 — both Evolve-reviewed papers are 2026 first-author arXiv preprints. The moderate score reflects a short publication history rather than any gap; sustained output beyond 2026 has not yet been observed and will re-rank this axis.
Reproducibility55/100
Reproducibility scores 55, computed as a mean of 55 over 2 scored papers, both rated medium readiness. Evolve's verdicts temper the headline numbers: MAGIC ranks first on 8 benchmarks but its single-run results need caution, while MACE's 81.11% versus SAGE's 78.97% rests on single-run evaluations, sub-0.1pp margins, and uncited baselines — methods look reportable, but claims await multi-run confirmation.
Research summary
Kairui Yang is a researcher at Peking University whose cataloged work concentrates entirely on LLM-based multi-agent systems. The profile is early-stage and recent: OpenAlex records 4 works in 2026 (countsByYear 2026: 4; recent3yWorks 4), and both Evolve-reviewed papers are first-author arXiv preprints from 2026. MAGIC constructs mixed-granularity agent graphs incrementally with dense-reward reinforcement learning, learning per-role atomic-or-group topologies and ranking first on 8 benchmarks — Evolve calls the idea sound but flags single-run results. MACE pairs agents with adaptive memory graphs in a co-evolution loop, reporting 81.11% versus SAGE's 78.97% across 8 benchmarks, though sub-0.1pp margins, single-run evaluations, and uncited baselines leave the SOTA claim unproven. Reflecting this concentration, the focus dimension scores a maximal 100 on a single recorded field, and topics cluster on AI-based problem solving and planning (3) and multi-agent systems and negotiation (2), with single mentions of topic modeling, semantic web, and robotics RL. Collaboration scores 51 across 12 distinct coauthors, including recurring partners such as Xunkai Li, Rong-Hua Li, Ziheng Yi, Zekai Chen, Minghao An, Zhanke Liu, and Guangyuan He. Reproducibility averages 55 over the 2 scored papers, both rated medium readiness. Citation metrics are recorded as h-index 0 and i10-index 0, so output and impact remain unscored pending accumulated evidence.
Collaboration pattern
Kairui Yang's cataloged work lists 12 distinct coauthors (collaboration score 51), with a clearly recurring core: Xunkai Li, Rong-Hua Li, Ziheng Yi, and Zekai Chen each appear twice in the frequency list, alongside Minghao An, Zhanke Liu, and Guangyuan He. The pattern suggests a small, stable in-house team behind the first-author multi-agent preprints rather than broad external networking, and both 2026 papers were written within this nucleus.
Research keywords
Highlights
MAGIC: RL-learned mixed-granularity agent graphs rank first on 8 benchmarks
MACE memory–agent co-evolution reports 81.11% vs SAGE's 78.97%
Timeline
- 2026Posts MAGIC on arXiv as first author: mixed-granularity agent graphs built incrementally with dense-reward reinforcement learning, ranked first on 8 benchmarks in Evolve's review.
- 2026Posts MACE on arXiv as first author: memory–agent co-evolution with adaptive memory graphs reporting 81.11% versus SAGE's 78.97% across 8 benchmarks.
- 2026OpenAlex countsByYear records 4 works, all in this single year, and recent3yWorks likewise equals 4 — marking the observable start of this profile.
- 2026Both preprints enter the Evolve catalog with medium reproducibility readiness and a mean reproducibility score of 55 across 2 scored papers.
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 are derived from OpenAlex only (identity: openalex); no ORCID is linked to this profile.
- h-index and i10-index are recorded as 0 with no citation history, so the impact axis is left unscored rather than estimated.
- Both cataloged papers are 2026 arXiv preprints; peer-review status and final venues are not captured in the available data.
- 'Kairui Yang' is a common Chinese name; this profile is name-resolved from public bibliographic data and should be verified against an ORCID or institutional page before citation.