Jian Zhu

University of British Columbia

Unclaimed · AI-structured profileOpenAlex

UBC researcher spanning biomedical retrieval, computational typology, and vision-language-action robot learning

papers here

reviewed by Evolve

first-author

corresponding

—

works · OpenAlex

—

h-index · OpenAlex

Bibliometric figures come from OpenAlex (2026-10-16) 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: only three 2026 arXiv papers are catalogued, too few to estimate publication volume, venue mix, or output trend. This axis would measure annual paper counts and venue distribution once fuller bibliographic records are linked.

Impact—

Not yet scored: no citation counts, h-index, or citation trajectories were available for any catalogued paper. This axis would measure citations and scholarly influence of the three 2026 papers once citation data are linked.

Focus0/100

The focus score is 0 because the three catalogued papers split evenly across three fields — biomedical NLP/information retrieval, computational linguistics (NLP), and robot learning/vision-language-action models — each accounting for exactly one third (0.333) of the record. OntologyBench sits in biomedical retrieval, LLM Agents as Computational Typologists in computational linguistics, and JEPA-WAM in robot learning. Breadth across communities, not concentration in one line of work, therefore defines this early profile.

Collaboration59/100

Collaboration scores 59 out of 100, based on 18 distinct coauthors in the record. Recurring partners include Xiao Yu Cindy Zhang, Wyeth Wasserman, Changbing Yang, Christopher Hammerly, Freda Shi, Tianbin Liu, Taiyi Su, Jianjun Zhang, Chong Ma, Zitai Huang, and Yi Xu, a roster spanning the biomedical, linguistics, and robotics strands of the work. Author-position data show a middle placement on JEPA-WAM and last-author placement on both 2026 benchmark papers, with none flagged corresponding.

Activity—

Not yet scored: all catalogued papers carry the same 2026 date, so publishing cadence, gaps, and recent-year momentum cannot be computed. This axis would measure annual research activity rhythm once a longer publication history is linked.

Reproducibility73/100

Reproducibility is the strongest measured axis, scoring 73 — the mean readiness across the three Evolve-reviewed papers. OntologyBench and LLM Agents as Computational Typologists are rated high readiness, while JEPA-WAM is rated low: its review notes no task-agnostic control, no significance tests, and no released code. The pattern points to a benchmark-building habit that supports replication, alongside one systems paper whose claims rest on untested comparisons.

Research summary

Jian Zhu is a researcher at the University of British Columbia whose visible work spans three fields in roughly equal measure: biomedical NLP and information retrieval, computational linguistics, and robot learning with vision-language-action models. The Evolve catalog holds three 2026 arXiv papers. In biomedical retrieval, OntologyBench asks whether dense retrievers can satisfy structured biomedical constraints over HPO and MONDO, finding that dense retrieval lags an ontology-aware reference on compositional tasks (0.313 vs 0.739 nDCG@10). In computational linguistics, LLM Agents as Computational Typologists tests whether LLM agents can code Grambank typological features from reference grammars and interlinear glossed text; the evaluation is careful and honest, but only three of five models beat the majority-vote baseline in the best setting. In robotics, JEPA-WAM augments World Action Models with text-to-image-generated, V-JEPA-encoded visual instructions, reporting strong real-robot wins over Fast-WAM. The record lists 18 distinct coauthors, yielding a mid-range collaboration score of 59, while the even three-way field split yields a focus score of 0. Reproducibility is the strongest measured dimension, with a mean readiness of 73 across the three papers — two benchmarks rated high, the robotics paper low. Output, impact, and activity cannot yet be scored, so this is an early-stage profile built solely from catalogued evidence.

Collaboration pattern

The collaboration score (59/100) rests on 18 distinct coauthors in the record. Recurring names include Xiao Yu Cindy Zhang, Wyeth Wasserman, Changbing Yang, Christopher Hammerly, Freda Shi, Tianbin Liu, Taiyi Su, Jianjun Zhang, Chong Ma, Zitai Huang, and Yi Xu — a mix consistent with the three-way spread across biomedical, linguistics, and robotics work. Zhu is a middle author on JEPA-WAM and last author on both 2026 benchmarks, none flagged corresponding.

Research keywords

dense retrievalbiomedical ontologiescompositional retrievalphenotype-disease matchingLLM agentslinguistic typologyreference grammarsinterlinear glossed text (IGT)vision-language-action modelsWorld Action Modelsinstruction followingtext-to-image augmentation

Highlights

OntologyBench exposes a structured-constraint gap in dense biomedical retrieval
The benchmark asks whether dense retrievers can satisfy structured biomedical constraints over HPO and MONDO. Dense retrieval reaches 0.313 nDCG@10 against 0.739 for the ontology-aware reference on compositional tasks. Evolve rates the benchmark high reproducibility readiness, though results are single-seed without significance tests.
LLM agents read reference grammars to code linguistic typology
The paper tests whether LLM agents can extract Grambank typological features from reference grammars and interlinear glossed text. Evaluation is careful and honest, but in the best setting only three of five models beat the majority-vote baseline. Reproducibility readiness is rated high.
JEPA-WAM injects generated visual instructions into World Action Models
The work augments World Action Models with text-to-image-generated instructions encoded by V-JEPA, reporting strong real-robot wins over Fast-WAM. Evolve finds the approach plausible but flags the absence of a task-agnostic control, significance tests, and released code, giving low reproducibility readiness.

Timeline

2026Publishes JEPA-WAM on arXiv, connecting T2I-generated, V-JEPA-encoded visual instructions to World Action Models with reported real-robot gains over Fast-WAM.
2026Releases OntologyBench, showing dense retrieval trails the ontology-aware reference on compositional biomedical retrieval (0.313 vs 0.739 nDCG@10).
2026Publishes LLM Agents as Computational Typologists, testing whether LLM agents can code Grambank typological features from reference grammars.

Papers in the Evolve catalog

Frequent co-authors

Xiao Yu Cindy Zhang×1 Wyeth Wasserman×1 Changbing Yang×1 Christopher Hammerly×1 Freda Shi×1 Tianbin Liu×1 Jian Zhu×1 Taiyi Su×1 Jianjun Zhang×1 Chong Ma×1 Zitai Huang×1 Yi Xu×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.

Jian Zhu is a common name; this profile is name-resolved from public bibliographic data and may conflate same-named authors (the coauthor list even contains a self-named entry).
Bibliometric scores derive from OpenAlex-based public data only; no ORCID or country field was available to verify identity.
Output, impact, and activity are unscored because the record holds only three 2026 arXiv papers; no citation, h-index, or venue-acceptance data were provided.
Reproducibility (73) covers only the three Evolve-reviewed papers; verdicts reflect Evolve's review assessments, not independent replication.