Zekai Chen

—

Unclaimed · AI-structured profileOpenAlex

Early-career researcher in graph-structured LLM multi-agent systems: topology learning, memory, and benchmarking

papers here

reviewed by Evolve

first-author

corresponding

—

works · OpenAlex

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—

No output score yet: the tracked record contains only three 2026 arXiv preprints, so total publication counts, venue quality, and productivity trends cannot be computed. This axis would measure cumulative output volume and its distribution over time and publication venues.

Impact—

No impact score yet: h-index and i10-index are recorded as 0, but all three papers are 2026 preprints too recent to have accrued citations. This axis would measure citation-based influence, including citation counts, h-index, and their growth over time.

Focus8/100

The focus score of 8 reflects an unusually concentrated agenda: per the field distribution, every catalog paper sits in machine learning on LLM-based multi-agent systems (66.7% systems, 33.3% evaluation), and the dominant topic, Multi-Agent Systems and Negotiation, covers all 3 papers. Side topics — topic modeling, NLP techniques, and reinforcement learning in robotics — appear only once each, functioning as supporting methods rather than separate research lines.

Collaboration54/100

With a collaboration score of 54, Chen works within a mid-sized, stable team: 14 distinct coauthors appear across the 3 catalog papers, all in middle author positions and none as corresponding author. The recurring core — Xunkai Li, Rong-Hua Li, and Kairui Yang, each listed twice among frequent collaborators — suggests a sustained group, complemented by Minghao An, Ziheng Yi, Kaixiang Zhang, Yuxuan Ba, and Zhanke Liu.

Activity40/100

The activity score of 40 rests on a very short tracked window: counts by year show 3 works, all in 2026, matching the recent3yWorks value of 3. All three are arXiv preprints released within the same year, so current output is concentrated rather than sustained; whether this pace continues cannot yet be judged from the available data.

Reproducibility55/100

Reproducibility averages 55 across the 3 scored papers, each rated medium readiness. Evolve's verdicts explain the mid-level score: MAGIC and MACE report single-run evaluations — with MACE's lead at 81.11% over SAGE's 78.97% resting on sub-0.1pp margins — while OpenMAS-GCom's intervention evidence covers only five of its 29 datasets.

Research summary

Zekai Chen is an early-stage researcher working almost exclusively on LLM-based multi-agent systems (MAS), with this profile name-resolved from OpenAlex bibliographic data. All three papers in the Evolve catalog appeared as arXiv preprints in 2026, and the field distribution is strikingly narrow: 66.7% of attributed work sits in LLM-based multi-agent systems and 33.3% in their evaluation, consistent with a focus score of just 8. The 2026 output concentrates on three graph-centric contributions: MAGIC learns mixed-granularity agent topologies with dense-reward reinforcement learning and ranks first on 8 benchmarks; MACE introduces memory-agent co-evolution over adaptive memory graphs, reporting 81.11% versus SAGE's 78.97%; and OpenMAS-GCom proposes a controlled-intervention diagnostic benchmark spanning 29 datasets. Evolve's reviews consistently rate reproducibility as medium (mean readiness 55 across 3 papers) and flag single-run evaluations and thin margins. Co-authorship data lists 14 distinct collaborators, with Xunkai Li, Rong-Hua Li, and Kairui Yang recurring; Chen sits in a middle author position on all three papers and has not yet held a corresponding role. Citation indicators (h-index 0, i10-index 0) reflect preprints too recent to accumulate citations, so the output and impact dimensions remain unscored.

Collaboration pattern

Chen's collaboration pattern shows a compact, recurring team: 14 distinct coauthors across 3 papers, with Xunkai Li, Rong-Hua Li, and Kairui Yang each appearing twice among frequent collaborators, alongside Minghao An, Ziheng Yi, Kaixiang Zhang, Yuxuan Ba, and Zhanke Liu. Chen holds middle author positions on all three papers and has not yet been corresponding author, consistent with a contributing member of an established multi-agent systems group rather than an independent lead.

Research keywords

LLM multi-agent systemsAgent topology generationGraph-enhanced multi-agent systemsAgent memory graphsReinforcement learningReward shapingExperience reuseBenchmark evaluationCommunication topologyRole assignmentControlled intervention analysisGraph construction policy

Highlights

MAGIC ranks first on 8 benchmarks with RL-learned agent topologies
Per Evolve's review, MAGIC learns per-role atomic-or-group multi-agent topologies through incremental construction with dense-reward, shaped reinforcement learning, ranking first on 8 benchmarks; the verdict calls the idea sound but cautions that single-run results need careful treatment.
MACE's memory-agent co-evolution reports 81.11% over SAGE's 78.97%
Evolve notes MACE's adaptive memory graphs let agents co-evolve with reusable experience, topping 8 benchmarks at 81.11% versus SAGE's 78.97% — but single-run evaluations, sub-0.1pp margins, and uncited baselines leave the state-of-the-art claim unproven.
OpenMAS-GCom diagnoses graph-enhanced MAS with controlled interventions
The benchmark claims controlled-intervention diagnosis of graph-enhanced MAS components across 29 datasets; Evolve rates the design rigorous but notes intervention evidence spans only five datasets, with medium reproducibility.

Timeline

2026Published the MAGIC preprint on arXiv, learning mixed-granularity agent graphs via incremental construction with dense-reward reinforcement learning.
2026Released MACE, introducing memory-agent co-evolution with adaptive memory graphs, reporting 81.11% versus SAGE's 78.97%.
2026Released OpenMAS-GCom, a controlled-intervention diagnostic benchmark for graph-enhanced multi-agent systems spanning 29 datasets.

Papers in the Evolve catalog

Frequent co-authors

Minghao An×3 Xunkai Li×3 Rong-Hua Li×3 Kairui Yang×2 Xunkai Li×2 Zekai Chen×2 Rong-Hua Li×2 Kairui Yang×2 Ziheng Yi×2 Kaixiang Zhang×1 Yuxuan Ba×1 Zhanke Liu×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.

Bibliometric data comes from OpenAlex only; no ORCID or institutional affiliation is available, and Zekai Chen is a common name, so this profile is name-resolved from public bibliographic records.
Output and impact dimensions are unscored: h-index and i10-index are recorded as 0 because all catalog papers are 2026 arXiv preprints without citation coverage yet.
All evidence is limited to three 2026 arXiv preprints reviewed by Evolve; no peer-reviewed venue, award, or career information is included in the data.
Reproducibility and collaboration metrics are computed over only 3 papers and 14 coauthors, so scores may shift substantially as the record grows.