Multimodal ML researcher building benchmarks and RL pipelines for audio-visual dialogue and user-intent understanding
2
papers here
2
reviewed by Evolve
1
first-author
0
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—
Not yet scored: the supplied record is too thin to compute an output axis. This dimension would measure total publication volume, venue placement, and the share of first- or corresponding-author papers across the indexed career.
Impact—
Not yet scored: no citation counts or influence metrics are available in the supplied data. This dimension would measure citations, downstream adoption of the two benchmarks, and real-world uptake of the training recipe.
Focus0/100
Focus scores 0/100. Within the reviewed catalog, the papers split exactly evenly: 50% in multimodal machine learning (speech and audio-visual interaction) and 50% in multimodal machine learning (speech and vision). The two sub-fields are adjacent — both center on audio-visual multimodal LLMs — so the zero reflects breadth across neighboring topics rather than scattered research; a two-paper catalog can also make the split look sharper than the underlying agenda.
Collaboration70/100
Collaboration scores 70/100, computed from 32 distinct co-authors in the record. Both 2026 papers are large-team efforts: the first-authored Omni Demand Understanding benchmark and the middle-authored OmniVChat systems paper each assemble sizable author lists. Frequent collaborators include Yunfei Chu, Jin Xu, Xie Chen, Qiuqiang Kong, Ziyang Ma, and Yuxuan Wang, consistent with an established seat inside a speech-and-multimodal research group rather than isolated authorship.
Activity—
Not yet scored: the record contains only 2026-dated papers, so publication cadence cannot be measured. This dimension would track year-by-year output, recent-productivity trends, and sustained engagement across the career timeline.
Reproducibility78/100
Reproducibility scores 78/100, the mean over 2 reviewed papers. The first-authored ODU-Bench rates high readiness: an explicit benchmark with a solid pipeline makes demand and intent inference directly checkable. OmniVChat rates medium readiness, reflecting judge bias and training trade-offs. Both papers ship public benchmarks, though reliance on LLM-judged metrics — 60% of the ODU score — keeps the axis short of the top band.
Research summary
Qi Chen is a researcher in multimodal machine learning, working where speech, audio, and vision meet large language models. The Evolve catalog holds two 2026 arXiv papers, split exactly evenly across two adjacent sub-fields: multimodal machine learning (speech and audio-visual interaction) and multimodal machine learning (speech and vision). The first-authored Omni Demand Understanding paper introduces ODU-Bench, which turns demand detection and contextual user-intent inference in multimodal interaction into an explicit benchmark, covering false-trigger cases; Evolve judges the pipeline solid, rates its reproducibility readiness high, and notes that an LLM-judged metric carries 60% of the score. The middle-authored OmniVChat paper contributes a synthetic data engine, a rubric-based benchmark, and reinforcement learning with rubric-based rewards for native audio-visual dialogue; gains transfer to real recordings, though judge bias and trade-offs remain, and reproducibility readiness is rated medium. The record lists 32 distinct co-authors, with frequent collaborators including Yunfei Chu, Jin Xu, Xie Chen, Qiuqiang Kong, Ziyang Ma, and Yuxuan Wang — a pattern consistent with large-team benchmark engineering in a speech-and-multimodal group. Scores computed from this data: collaboration 70/100, reproducibility 78/100, and focus 0/100, the last reflecting the even field split. Output, impact, and activity axes lack sufficient data and are left unscored. This profile is name-resolved from public bibliographic data only.
Collaboration pattern
The record lists 32 distinct co-authors, driving a collaboration score of 70/100. Both 2026 papers are multi-author team efforts — a first-authored benchmark and a middle-authored systems paper. The most frequent collaborators include Yunfei Chu, Haolin He, Wen Huang, Qiuqiang Kong, Jin Xu, Xie Chen, Yifan Yang, Zihan Liu, Yuxuan Wang, Ziyang Ma, and Ruiyang Xu, a roster consistent with a speech-and-multimodal lab running large-scale benchmark and data-engine projects.
Research keywords
Highlights
ODU-Bench: first-authored benchmark for demand and intent inference
OmniVChat: synthetic data engine and rubric-reward RL for native audio-visual dialogue
A recurring caveat: dependence on LLM-judged evaluation
Both reviewed 2026 papers lean on model-based evaluation and both are flagged for it: ODU-Bench's LLM-judged metric carries 60% of its score, and OmniVChat retains judge bias after training. The repeated caveat points to evaluation reliability as the open problem in this benchmark engineering agenda.
Timeline
- 2026Publishes ODU-Bench as first author — a benchmark for contextual user-intent inference in multimodal interaction (arXiv, speech and audio-visual interaction field).
- 2026Contributes as a middle author to OmniVChat, which combines synthetic data generation, a rubric benchmark, and reinforcement learning for native audio-visual dialogue (arXiv, speech and vision field).
- 2026Both papers enter the Evolve catalog; reproducibility across the two reviewed papers averages 78, with ODU-Bench rated high and OmniVChat rated medium.
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 and radar scores are computed from OpenAlex-derived public bibliographic data only; citation and venue coverage may be incomplete.
- "Qi Chen" is a common name; this profile is name-resolved from public bibliographic data and could merge or miss records of different researchers sharing the name.
- No institution, country, or ORCID is attached to this record, so affiliation and career context are absent.
- Output, impact, and activity are unscored because the supplied data covers only two 2026 papers with no citation counts.