Yunfei Chu

—

Unclaimed · AI-structured profilename-matched identity

Multimodal machine learning researcher: benchmarks and training for audio-visual dialogue with omni-modal LLMs

papers here

reviewed by Evolve

first-author

corresponding

—

works · OpenAlex

—

h-index · OpenAlex

Bibliometric figures come from OpenAlex (2026-10-14) 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—

Not yet scored: the indexed snapshot provides no publication counts or yearly output volumes for this name record. This axis would measure total production and per-year paper counts across the cataloged bibliography, here limited to the two Evolve-reviewed 2026 preprints.

Impact—

Not yet scored: citation counts, h-index, or other influence metrics are absent from this snapshot. The axis would measure how often the reviewed papers are cited and how widely the audio-visual benchmark and training work has been adopted.

Focus0/100

Scores 0 on the 0-100 focus scale, driven by an exactly even field distribution: 50% multimodal machine learning for speech and audio-visual interaction, 50% for speech and vision. Both cataloged papers sit in these adjacent sub-areas, so the minimum score reflects breadth across two closely related branches of one broader area — audio-centric multimodal LLM research — rather than scattering across unrelated fields.

Collaboration68/100

Scores 68 on the 0-100 scale, computed from 29 distinct coauthors in the indexed record, of whom 12 recur as frequent collaborators — including Qi Chen, Haolin He, Jin Xu, Xie Chen, Qiuqiang Kong, Yifan Yang, Ziyang Ma, and Yuxuan Wang. Both 2026 papers place the researcher in a middle author position without corresponding-author role, consistent with contributing specialist work inside larger multi-author teams.

Activity—

Not yet scored: recent publication cadence cannot be computed from the snapshot. This axis would measure publishing activity in the latest one-to-two-year window, such as new preprints and submissions beyond the two cataloged 2026 papers.

Reproducibility78/100

Scores 78, the mean readiness across the 2 Evolve-reviewed papers. Omni Demand Understanding rates high: an explicit benchmark with a solid pipeline, though an LLM-judged metric carries 60% of its score. OmniVChat rates medium: its synthetic data engine, rubric benchmark, and reinforcement-learning pipeline show gains that transfer to real recordings, but judge bias and methodological trade-offs remain open issues.

Research summary

Yunfei Chu is a researcher in multimodal machine learning whose cataloged work centers on audio-visual interaction for omni-modal large language models. The indexed field profile splits exactly evenly (50/50) between speech-and-audio-visual interaction and speech-and-vision, and both Evolve-reviewed papers — OmniVChat and Omni Demand Understanding (ODU-Bench) — appeared as arXiv preprints in 2026 with the researcher in a middle, non-corresponding author position. OmniVChat contributes a multi-agent synthetic data engine, a rubric-based benchmark, and reinforcement learning with rubric-based rewards for native audio-visual dialogue; Evolve notes the gains transfer to real recordings while judge bias and trade-offs remain. ODU-Bench turns demand detection and contextual user-intent inference into an explicit benchmark covering audio-visual interaction and false-trigger cases, with a solid pipeline whose LLM-judged metric carries 60% of the score. The collaboration dimension scores 68 (0-100), built on 29 distinct coauthors and a recurring core of 12 frequent collaborators; reproducibility scores 78, the mean across the 2 reviewed papers, with ODU-Bench rated high readiness and OmniVChat medium. Output, impact, and activity dimensions lack underlying bibliometric data and are therefore unscored. Institution, country, and ORCID are absent from the record, so this profile is name-resolved from public bibliographic data.

Collaboration pattern

The indexed record shows 29 distinct coauthors, with a recurring core of 12 frequent collaborators — Qi Chen, Haolin He, Wen Huang, Qiuqiang Kong, Jin Xu, Xie Chen, Yifan Yang, Zihan Liu, Yuxuan Wang, Ziyang Ma, Ruiyang Xu, and Meng Gao. Both 2026 papers list Yunfei Chu in a middle, non-corresponding position, indicating work within large multi-author teams building complete systems (data engine, benchmark, training) rather than solo or corresponding-author projects.

Research keywords

audio-visual dialogueomni-modal large language modelssynthetic data generationmulti-agent data enginereinforcement learningrubric-based rewardbenchmarkdemand understandingintent inferencefalse trigger

Highlights

Synthetic data, rubric benchmark, and RL for native audio-visual dialogue
OmniVChat couples a multi-agent synthetic data engine with a rubric-based benchmark and reinforcement learning using rubric-based rewards; Evolve notes the resulting gains transfer to real recordings, while judge bias and capability trade-offs remain.
Making demand understanding and contextual intent inference an explicit benchmark
ODU-Bench formalizes demand detection and contextual user-intent inference in multimodal interaction, covering audio-visual interaction and false-trigger cases; the pipeline is solid and reproducibility readiness is rated high, though an LLM-judged metric carries 60% of the score.
Benchmarks plus training data: a paired 2026 contribution
Both cataloged papers appeared in 2026 and attack the same problem from two sides — ODU-Bench measures what omni-modal models understand, while OmniVChat shows how to synthesize data and train for it — within a collaboration record spanning 29 distinct coauthors.

Timeline

2026Co-authored OmniVChat (arXiv, middle author position), a synthetic data engine, rubric benchmark, and reinforcement-learning recipe for native audio-visual dialogue
2026Co-authored ODU-Bench (arXiv, middle author position), an omni demand understanding benchmark with high reproducibility readiness
2026Published both cataloged papers in the same year, concentrating on benchmarks and training data for audio-visual interaction in multimodal LLMs

Papers in the Evolve catalog

Frequent co-authors

Qi Chen×2 Haolin He×2 Wen Huang×2 Qiuqiang Kong×2 Jin Xu×2 Xie Chen×2 Yifan Yang×1 Zihan Liu×1 Yuxuan Wang×1 Ziyang Ma×1 Ruiyang Xu×1 Meng Gao×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 coauthor counts derive from public bibliographic indexing (OpenAlex-style records) only; no citation counts, h-index, or venue-level data are included.
Institution, country, and ORCID are absent from this record; the profile is name-resolved from public bibliographic data and may merge or split namesakes.
Only the 2 Evolve-reviewed 2026 arXiv preprints are itemized here; the full record behind the collaboration and field metrics is not listed.
Output, impact, and activity dimensions are unscored because the underlying metrics are missing from the snapshot, not because they are zero.