Michigan researcher in multimodal reasoning and long-horizon agent evaluation, in a steep 2025–26 publishing ramp
2
papers here
2
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
3
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—
Output cannot be scored yet: this profile lacks a verified total-publication list and venue-normalized output record. Once a complete publication set is linked, this axis would measure annual output volume, venue quality, and authorship roles across the career.
Impact—
Impact cannot be scored yet: reliable citation counts and downstream uptake are not available for this record. With a complete citation graph, this axis would measure citations per paper, referencing works, and influence on subsequent research.
Focus0/100
Focus scores 0 because activity is evenly split — the field distribution is 50% artificial intelligence (long-context multimodal evaluation) and 50% machine learning (multimodal vision-language models) — and indexed topics scatter across five clusters: multimodal ML (9), robot manipulation (7), planning (5), topic modeling (3), and RL in robotics (3). Current work runs parallel threads rather than one concentrated line.
Collaboration62/100
Collaboration scores 62, built on 22 distinct coauthors in the indexed record. The recurring cluster includes Yichi Zhang (listed three times), Jiayuan Mao, Ziqiao Ma, Run Peng, Jing Ding, Yinpei Dai, Zengqing Wu, and Yao Fu, spanning both vision-language reasoning and embodied-agent projects; the two Evolve-reviewed 2026 papers were written in middle- and last-author positions within such teams.
Activity79/100
Activity scores 79: the record indexes 14 works in the most recent three-year window, with output accelerating sharply — 1 work in 2023, 3 in 2025, and 11 in 2026. This is a steep recent ramp, though the short coverage window limits how much of the longer career trajectory the score reflects.
Reproducibility55/100
Reproducibility averages 55 across the 2 papers with review data, both rated medium readiness. Evolve notes concrete gaps: the SVG-overlay reasoning paper reports roughly 6.5–10 weighted-point gains over SFT on geometry but no significance tests or stated code release, and SIMLIFE's headline numbers depend on oracle-filtered logs and LLM video summaries — rigorous design, partly closed evaluation.
Research summary
This profile covers Joyce Chai of the University of Michigan, resolved from public OpenAlex bibliographic data. The indexed record is recent and accelerating: 14 works appear in the last three years, with yearly counts rising from 1 in 2023 to 3 in 2025 and 11 in 2026. Activity splits evenly between machine learning on multimodal vision-language models and artificial intelligence on long-context multimodal evaluation, with topic clusters led by multimodal machine learning applications (9) and robot manipulation and learning (7), followed by AI-based planning (5), topic modeling (3), and reinforcement learning in robotics (3). Evolve has reviewed two 2026 arXiv preprints. 'Multimodal Thinking with Renderable Programs' (middle author) tunes vision-language models to sketch SVG overlays while reasoning, gaining roughly 6.5–10 weighted points over SFT on geometry. 'SIMLIFE' (last author) introduces a Sims 4 benchmark probing latent behavioral-rule understanding over long horizons. Both are rated medium for reproducibility, averaging 55. The record lists 22 distinct coauthors, with recurring collaborators including Yichi Zhang, Jiayuan Mao, Ziqiao Ma, Run Peng, Yao Fu, and Zengqing Wu. Citation indicators are thin — h-index 3, i10-index 1 — likely reflecting an incomplete author record rather than mature impact evidence.
Collaboration pattern
Collaboration scores 62, with 22 distinct coauthors in the indexed record. Work clusters around a recurring group — Yichi Zhang appears three times in the frequent-collaborator list, alongside Jiayuan Mao, Ziqiao Ma, Run Peng, Jing Ding, Yinpei Dai, Zengqing Wu, and Yao Fu — spanning both vision-language reasoning and embodied-agent projects. The two Evolve-reviewed 2026 papers were written in middle and last authorship positions within such teams.
Research keywords
Highlights
Sketch-to-think: SVG overlays as visual chain-of-thought
SIMLIFE: probing long-horizon behavioral-rule understanding
Timeline
- 2023Earliest indexed activity in this OpenAlex record: 1 work listed for the year.
- 20253 works indexed — the start of a visible ramp in publishing output.
- 202611 works indexed in a single year, including two arXiv preprints reviewed by Evolve: 'Multimodal Thinking with Renderable Programs' (middle author) and 'SIMLIFE' (last author).
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 name-resolved from a single OpenAlex record (identity: openalex) with no ORCID linkage; the thin citation indicators (h-index 3, i10-index 1) may reflect an incomplete author record rather than the researcher's full output.
- The coverage window is short: the record indexes works only in 2023, 2025, and 2026, so earlier-career output is not represented.
- Both reviewed papers are recent arXiv preprints rated medium for reproducibility (mean 55 over 2 papers); per Evolve review, neither states significance testing or code release.
- The frequent-coauthor list contains the subject's own name and duplicate entries (Yichi Zhang three times), an extraction artifact; collaboration figures are interpreted accordingly.