Cheng Chen

Tongji University

Unclaimed · AI-structured profileOpenAlex

Cheng Chen (Tongji University): LLM-agent social simulation audits meet multimodal audio-visual learning

papers here

reviewed by Evolve

first-author

corresponding

—

works · OpenAlex

—

h-index · OpenAlex

Bibliometric figures come from OpenAlex (2026-10-13) 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 publication-volume data is available in the current snapshot, so this axis cannot be scored. Once computed, it would measure total paper count, yearly output, and authorship roles over time.

Impact—

Citation-based influence cannot be assessed: no citation counts, h-index, or venue metrics exist in the underlying data. This axis would measure citation accumulation and scholarly uptake of the reviewed work.

Focus0/100

The focus score is 0, and the raw field distribution explains why: activity splits evenly, 0.5 to multimedia / multimodal machine learning and 0.5 to AI evaluation / multi-agent systems (LLM social simulation). Each of the two 2026 arXiv papers sits in a different field — a theory-of-mind audit of multi-agent social simulation, and open-vocabulary audio-visual localization via knowledge distillation — indicating a dual-track rather than concentrated agenda at this stage.

Collaboration44/100

Collaboration scores 44 on a 0-100 scale, backed by 8 distinct coauthors across the two cataloged papers. Named frequent collaborators include Yi Xu, Wenzhuo Lei, Cong Li, Thomas Fung, Alex Rossi, and Yi Li. On both 2026 arXiv papers Chen appears as a middle author without corresponding authorship, suggesting a contributing role within group-led projects rather than an independent, self-led collaboration network so far.

Activity—

Temporal productivity data is insufficient: both cataloged papers are dated 2026, leaving no multi-year record to gauge cadence. This axis would measure publishing continuity and recent output rhythm.

Reproducibility55/100

Reproducibility averages 55 across the 2 assessed papers, both tagged medium reproducibility readiness. The distillation paper is described as well ablated, with 0.816 AP gains reported, though its evidence rests mainly on one benchmark; the theory-of-mind audit is judged sound but carries a single-reader caveat. Together these verdicts support a middling score: solid evaluation hygiene constrained by narrow benchmark coverage and audit limitations.

Research summary

Cheng Chen, affiliated with Tongji University, is an early-stage researcher whose cataloged work spans two fields in near-even measure: multimedia / multimodal machine learning and AI evaluation / multi-agent systems centered on LLM social simulation (field distribution 0.5 / 0.5). The Evolve catalog holds two 2026 arXiv papers, on both of which Chen is a middle-author contributor rather than first or corresponding author. The first, 'Mind or Message? Auditing Theory of Mind in Multi-Agent Social Simulation', asks whether LLM agents in multi-issue negotiation actually model their interlocutors; Evolve's verdict finds the agents fluent but projection-driven and blind to the Pareto frontier, judging the audit sound while noting a single-reader caveat. The second, 'If You Hear It, Help Find It', proposes OV-OrthKD, a reliability-aware asymmetric distillation method for open-vocabulary audio-visual event localization, delivering well-ablated 0.816 AP gains whose evidence rests mainly on a single benchmark. Quantitative depth remains limited: the output, impact, and activity axes lack sufficient underlying data, so no publication totals, citation counts, or multi-year cadence can be reported. Collaboration scores 44 with 8 distinct coauthors, including frequent collaborators Yi Xu, Wenzhuo Lei, Cong Li, Thomas Fung, Alex Rossi, and Yi Li. Reproducibility averages 55 across the two assessed papers, both rated medium readiness — solid ablation practice constrained by narrow benchmark coverage and audit caveats.

Collaboration pattern

Chen works in small team settings: 8 distinct coauthors appear across the two cataloged 2026 papers, with Yi Xu, Wenzhuo Lei, Cong Li, Thomas Fung, Alex Rossi, and Yi Li among the most frequent. Holding middle-author, non-corresponding positions on both papers, Chen currently functions as a contributing collaborator within group projects; the moderate collaboration score of 44 reflects a small but repeatable network rather than a broad one.

Research keywords

theory of mindLLM agentssocial simulationmulti-issue negotiationPareto frontieregocentric projectionopen-vocabulary learningaudio-visual event localizationknowledge distillationmultimodal fusiontemporal groundingteacher reliability

Highlights

Auditing Theory of Mind in LLM Social Simulation
In a 2026 arXiv audit of multi-issue negotiation among LLM agents, Evolve finds that agents track messages rather than minds: responses are fluent but driven by egocentric projection and blind to the Pareto frontier. The audit itself is judged sound, with a single-reader caveat noted.
Reliability-Aware Distillation for Audio-Visual Localization
The 2026 OV-OrthKD paper proposes reliability-aware asymmetric knowledge distillation for open-vocabulary audio-visual event localization. Evolve credits well-ablated experiments with 0.816 AP gains, while cautioning that the evidence rests mainly on a single benchmark.
A Dual-Field, Evaluation-First 2026 Output
Both 2026 papers emphasize measurement over novelty claims alone: one audits whether agents genuinely model minds, the other adds ablations and teacher-reliability controls to distillation. The field split is even (0.5 / 0.5), marking an early-career portfolio spanning multimodal learning and agent evaluation.

Timeline

2026Publishes 'Mind or Message?', an arXiv audit of theory of mind in multi-agent LLM social simulation, as a middle author.
2026Publishes the OV-OrthKD paper on orthogonal knowledge distillation for open-vocabulary audio-visual event localization on arXiv.
2026The two releases establish an even 0.5 / 0.5 split between multimodal machine learning and LLM-agent evaluation, with 8 distinct coauthors recorded.

Papers in the Evolve catalog

Frequent co-authors

Yi Xu×1 Wenzhuo Lei×1 Cong Li×1 Thomas Fung×1 Alex Rossi×1 Yi Li×1 Cheng Chen×1 Yi Li×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 scores and counts come from OpenAlex-derived bibliometric data; only two 2026 arXiv preprints underlie this profile.
'Cheng Chen' is a common name; identity is name-resolved from public bibliographic data, and the co-author list contains repeated entries ('Cheng Chen' and 'Yi Li' twice), indicating record-level artifacts.
Output, impact, and activity axes lack underlying data — no citation counts, publication totals, or multi-year timeline; no ORCID or country is recorded to further confirm identity.
The reproducibility mean (55) rests on just 2 assessed papers, and Evolve verdicts cover only these two works; findings may not generalize to the full record.