Multimodal ML researcher in audio-visual dialogue and efficient video MLLMs, with two 2026 arXiv preprints.
2
papers here
2
reviewed by Evolve
0
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—
No output score yet: the source data provides no publication counts, venue list, or total record volume for this researcher. This axis would measure the size and consistency of the publication record, including the balance between preprints and peer-reviewed output.
Impact—
No impact score yet: citation counts, h-index, and per-paper influence are absent from the input data. This axis would measure how widely the work is read, cited, and built upon across the multimodal-learning community.
Focus0/100
The focus dimension scores 0, driven by an even field distribution: 50% multimodal machine learning (efficient video understanding) and 50% multimodal machine learning (speech and vision). In practice the two cataloged 2026 preprints sit in adjacent subfields — audio-visual dialogue and efficient long-video inference — so the score reflects breadth across a young two-paper sample rather than confirmed scatter; both share the same multimodal-LLM core.
Collaboration64/100
The collaboration dimension scores 64, anchored on 24 distinct coauthors recorded for this researcher. The recurring list includes Haoyu Guo, Junlin Lv, Mingjun Xiao, S Kevin Zhou, Xike Xie, Muzhi Zhu, and Zheqi Dai, and both cataloged papers are middle-author, non-corresponding contributions — a pattern of team-based work inside sizeable multimodal-LLM groups rather than solo or led projects. One listed coauthor duplicates the researcher's own name, likely a disambiguation artifact.
Activity—
No activity score yet: the data lacks a multi-year publication cadence, so recent output rate and research continuity cannot be measured. This axis would track how frequently new preprints and papers appear over successive years for this researcher.
Reproducibility55/100
The reproducibility dimension scores 55, computed as a mean readiness of 55 across 2 reviewed papers, each rated medium by Evolve's assessment. OmniVChat ships a synthetic data engine, a rubric benchmark, and RL training, but the review flags judge bias; VideoMM reports a 6.13x speedup (+7.4%) on LongVideoBench while carrying noted numeric inconsistencies and no error bars — substantive results, but with verification caveats.
Research summary
Yuan Feng is an early-stage researcher in multimodal machine learning whose public bibliographic footprint, name-resolved by Evolve from public bibliographic data, currently centers on two 2026 arXiv preprints at the intersection of speech–vision dialogue and efficient video understanding. The first, OmniVChat, builds a full stack for native audio-visual dialogue — a multi-agent synthetic data engine, a rubric-based benchmark, and reinforcement learning with rubric rewards — with gains that transfer to real recordings, though the review flags judge bias and remaining trade-offs. The second, VideoMM, targets efficiency in video multimodal LLMs through adaptive macro-micro, coarse-to-fine inference with visual token reduction and early exit, reporting a 6.13x speedup (+7.4%) on LongVideoBench, tempered by numeric inconsistencies and the absence of error bars. The field distribution is evenly split (50/50) between the two subfields, yielding a focus score of 0 — breadth over concentration at this stage. Collaboration is the strongest measured axis at 64, with 24 distinct coauthors including Haoyu Guo, Junlin Lv, Mingjun Xiao, S Kevin Zhou, and Xike Xie; both papers are middle-author, non-corresponding contributions, indicating a contributor role within larger teams. Reproducibility readiness averages 55 (medium) across the two papers. No institution, country, ORCID, citation, or publication-count data is available, so output, impact, and activity remain unscored, and this profile should be read as a young, fast-moving snapshot rather than a settled record.
Collaboration pattern
With 24 distinct coauthors recorded and a collaboration score of 64, Yuan Feng works in sizeable team settings: both 2026 papers are middle-author, non-corresponding contributions. Recurring names include Haoyu Guo, Junlin Lv, Mingjun Xiao, S Kevin Zhou, Xike Xie, Muzhi Zhu, and Zheqi Dai, consistent with large multimodal-LLM project groups. The coauthor list also contains an entry matching the researcher's own name, which Evolve treats as a likely name-disambiguation artifact rather than a separate collaborator.
Research keywords
Highlights
OmniVChat: full-stack recipe for native audio-visual dialogue
VideoMM: adaptive macro-micro inference speeds long-video MLLMs
Timeline
- 2026OmniVChat preprint appears on arXiv, introducing a synthetic data engine, a rubric benchmark, and reinforcement learning training for native audio-visual dialogue.
- 2026VideoMM preprint appears on arXiv, proposing adaptive macro-micro inference for efficient video MLLMs with a reported 6.13x speedup (+7.4%) on LongVideoBench.
- 2026Both cataloged papers land in the same year across two adjacent multimodal subfields, shaping a breadth-oriented early profile built on 24 distinct coauthors.
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 are derived from public bibliographic data only; no institution, country, or ORCID is attached, and Yuan Feng is a common name, so this profile is name-resolved and may merge or split distinct researchers.
- Output, impact, and activity dimensions lack underlying data (no publication counts, citations, or multi-year cadence) and are intentionally left unscored.
- The profile rests on two 2026 arXiv preprints with middle-author, non-corresponding positions — an early-stage snapshot that may not reflect the full publication record.
- The frequent-coauthor list contains an entry identical to the profile name (Yuan Feng), likely a name-disambiguation artifact in the source data.