Xuanjing Huang

Fudan University

Unclaimed · AI-structured profileOpenAlex

Fudan researcher spanning LLM-based social simulation, automated LLM fine-tuning, and multimodal LLM post-training

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 data is not yet available. This axis would measure publication volume and productivity — total paper counts, yearly tempo, and venue spread — which cannot be computed from the four 2026 arXiv entries currently in the catalog.

Impact—

Impact data is not yet available. This axis would measure citation-based influence — citation counts, h-index, and citation velocity — which require bibliometric sources beyond the current snapshot.

Focus0/100

Focus scores 0 because this researcher's catalogued work splits evenly across three fields — LLM-based social simulation, automated LLM fine-tuning with LLM agents, and multimodal LLM post-training — each holding 33.3% of the field distribution. The raw data suggests parallel, methodologically linked LLM threads rather than a single dominant specialty.

Collaboration82/100

Collaboration scores 82, backed by 60 distinct co-authors in the raw data. Twelve recur frequently, including Tao Gui, Qi Zhang, Xinnong Zhang, Shaofan Liu, Honglin Guo, Xinyi Mou, and Jiayu Lin, indicating a stable core group. All three assessed 2026 papers list this researcher as last author, with SocioVerse2 in the middle position — a pattern consistent with a senior role in team-scale projects.

Activity—

Activity data is not yet available. This axis would measure recent publishing rhythm — papers per year, gaps between outputs, and current momentum — but the four catalogued entries all date to 2026, a window too narrow to score.

Reproducibility55/100

Reproducibility scores 55, the mean readiness across three assessed papers, each rated medium. The evidence is mixed: SAGE reports nine-task results from only three runs without significance tests, ReDraft's headline metrics slightly clash with its own Table 4, and SocioVerse2's validation repackages the authors' companion studies — replication looks feasible but needs care.

Research summary

Xuanjing Huang (Fudan University) is a full-profile researcher whose catalogued work splits evenly into three LLM research threads: LLM-based social simulation in computational social science, automated LLM fine-tuning with LLM agents, and multimodal LLM post-training, each holding 33.3% of the field distribution. The Evolve catalog holds four 2026 arXiv entries. SocioVerse2 proposes a longitudinal social simulation framework under a human-AI co-evolutionary paradigm, evaluated across seven cases. SAGE makes automated fine-tuning cumulative through a strategy experience repository, lifting Avg@3 gains from 3.2% to 15.6% across nine tasks. ReDraft addresses catastrophic forgetting in vision-language cold-start post-training via self-revised verified targets. A fourth paper, "Agents in the Large," presents a perception-centered architecture for persistent agents. Collaboration is the standout dimension (score 82), grounded in 60 distinct co-authors and a frequent core of twelve including Tao Gui, Qi Zhang, and Xinnong Zhang. Reproducibility averages 55 across three assessed papers, each rated medium; Evolve notes SAGE rests on three runs without significance tests and ReDraft's headline metrics slightly clash with its own Table 4. Focus scores 0, reflecting an evenly split three-way portfolio rather than one dominant field. Output, impact, and activity axes lack data and remain unscored.

Collaboration pattern

Collaboration is this profile's strongest axis (score 82), with 60 distinct co-authors in the raw data. A recurring core of twelve — Tao Gui, Qi Zhang, Xinnong Zhang, Shaofan Liu, Honglin Guo, Xinyi Mou, Jiayu Lin, Jia Wang, Yixu Huang, Yingqian Wu, Jingcong Liang, and Shijun Lei — suggests a stable lab group. Authorship placement (last position on three catalogued 2026 papers, middle on SocioVerse2) is consistent with a senior authorship role in team-scale projects.

Research keywords

LLM-based social simulationagent-based modelingLLM agentshuman-AI collaborationcounterfactual interventionautomated LLM fine-tuningMonte Carlo Tree Searchexperience reusecontinual post-trainingcatastrophic forgettingvision-language modelson-policy distillation

Highlights

SocioVerse2: Human-AI Co-evolution for Longitudinal Social Simulation
SocioVerse2 frames long-horizon social simulation as a human-AI co-evolutionary process, supporting counterfactual intervention and backed by a solid seven-case evaluation; Evolve notes that its validation largely repackages the authors' companion studies.
SAGE: Experience Accumulation for Automated LLM Fine-Tuning
SAGE turns automated fine-tuning into a cumulative process via a strategy experience repository, lifting Avg@3 gains from 3.2% to 15.6% across nine tasks; Evolve flags that results rest on three runs with no significance tests.
ReDraft: Reference-Driven Revision for Continual VLM Post-Training
ReDraft counters catastrophic forgetting in vision-language cold-start post-training by self-revising verified targets instead of pure distillation; Evolve credits strong results and analysis but notes headline metrics slightly clash with the paper's own Table 4.

Timeline

2026Published SocioVerse2, a longitudinal dynamic social simulation framework under a human-AI co-evolutionary paradigm (middle authorship).
2026Published ReDraft, reference-driven revision for continual vision-language model post-training (last authorship).
2026Published SAGE on strategy accumulation and guided execution for automated LLM fine-tuning (last authorship).
2026Appeared on "Agents in the Large," a perception-centered architecture for persistent agents (last authorship).

Papers in the Evolve catalog

Frequent co-authors

Tao Gui×3 Shaofan Liu×2 Honglin Guo×2 Qi Zhang×2 Xinnong Zhang×1 Jiayu Lin×1 Jia Wang×1 Yixu Huang×1 Xinyi Mou×1 Yingqian Wu×1 Jingcong Liang×1 Shijun Lei×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.

Dimension scores come from the Evolve bibliometric snapshot only; output, impact, and activity axes lack sufficient data and are unscored.
All four catalogued papers are 2026 arXiv preprints; no peer-reviewed venues or earlier years are represented in this snapshot.
Only three papers have Evolve verdicts and reproducibility assessments (all medium); "Agents in the Large" lacks field, keywords, verdict, and reproducibility data.
No ORCID or country is provided; the profile is name-resolved from public bibliographic data, so identity should be confirmed before citing.