University of Liverpool researcher: cost-efficient LLM multi-agent debate and multimodal evidence-conflict evaluation
2
papers here
2
reviewed by Evolve
1
first-author
0
corresponding
—
works · OpenAlex
—
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-02) 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—
The output axis has no underlying data in this snapshot: no publication counts, venue history, or yearly output volume are available beyond the two reviewed arXiv preprints, so this dimension measures scholarly productivity and is left unscored rather than estimated.
Impact—
No citation counts, h-index, or influence metrics exist for this profile yet; the impact axis would measure how widely the two 2026 preprints are cited and adopted, so it remains unscored rather than guessed from thin evidence.
Focus0/100
Focus scores 0 on the 0-100 radar, reflecting the field distribution rather than absence of direction: work splits exactly 0.5/0.5 between LLM multi-agent systems and multimodal LLM evaluation. In practice the two lines are thematically adjacent — both probe how LLM-based systems process and weigh information — so the even split reads as an early, two-track agenda built from only two cataloged papers rather than as scattered interest.
Collaboration32/100
Collaboration scores 32 out of 100 with only 4 distinct coauthors across the record, marking compact, small-team research. Zhuoyun Li, Xiaowei Huang, and Yi Dong recur as frequent collaborators, consistent with a small, stable group; Wang appears as first author on the multi-agent debate paper and in a middle position on the multimodal evaluation paper, suggesting a hands-on contributing role within that team.
Activity—
The activity axis is not scored: this snapshot contains no publication-frequency, submission, or per-year productivity series beyond two 2026 preprints. It would measure research tempo — how steadily new work appears — and the current data are too thin to estimate it.
Reproducibility38/100
Reproducibility scores 38, the mean readiness across the 2 papers with reproducibility assessments. The multi-agent debate paper is rated medium: its cost-efficiency protocol is describable, but Evolve notes no code is released and accuracy gains fall within seed noise. The evidence-order paper is rated low, so independent replication of either result currently depends on the authors' descriptions rather than released artifacts.
Research summary
Boxuan Wang is a researcher at the University of Liverpool working at the intersection of LLM multi-agent systems and multimodal LLM evaluation — the two fields split Evolve's field distribution for this profile evenly at 0.5 each. The cataloged record consists of two 2026 arXiv preprints. The first, a first-author paper, asks whether sparse multi-agent debate needs complex topology control at all, proposing distinct-peer random routing with an early (round-3) stopping rule; Evolve's review finds the cost-efficiency story solid while noting that accuracy gains fall within seed noise and that no code is released. The second, a middle-author contribution, shows that reordering the same evidence shifts multimodal judgments toward image and speech inputs — a well-controlled empirical claim, though the term 'noncommutativity' extends beyond the two-element setting tested. Collaboration is compact: four distinct coauthors, with Zhuoyun Li, Xiaowei Huang, and Yi Dong recurring across the record. Bibliometric coverage is narrow — output, impact, and activity axes lack the underlying data to score, and citation counts, h-index, or publication-volume figures are not available in this snapshot. The profile is therefore best read as an early-stage, arXiv-centered view of a researcher whose published work so far concentrates on making LLM-agent communication cheaper and on stress-testing how multimodal models weigh conflicting evidence.
Collaboration pattern
Collaboration is compact: 4 distinct coauthors underlie the radar's 32/100 collaboration score, with Zhuoyun Li, Xiaowei Huang, and Yi Dong recurring as frequent collaborators. Wang took first-author responsibility on the multi-agent debate paper and a middle authorship position on the multimodal evaluation study, a pattern consistent with contributing deeply inside one small, stable team rather than across many ad-hoc partnerships; no large-scale or cross-institution collaboration is visible in this record.
Research keywords
Highlights
Random distinct-peer routing rivals complex topology control in sparse multi-agent debate
Same evidence, different judgments: order effects in multimodal LLM evaluation
Timeline
- 2026First-author arXiv preprint on distinct-peer random routing for cost-efficient sparse multi-agent debate.
- 2026Co-authors (middle position) a multimodal evaluation study showing that evidence order shifts vision/speech-text conflict judgments.
- 2026Both preprints enter the Evolve catalog with reproducibility reviews (medium and low readiness).
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.
- Radar scores are computed by Evolve from a limited bibliometric snapshot; only two reviewed papers underlie the field, collaboration, and reproducibility figures.
- Output, impact, and activity dimensions lack underlying data (no citation counts, h-index, or publication-volume series) and are left unscored.
- Both cataloged papers are 2026 arXiv preprints; peer-review status and final venues are not indicated in the snapshot.
- No ORCID or country is attached to this record; the profile is name-resolved from public bibliographic data and may merge or split same-name authors.