UC Berkeley NLP researcher working on LLM-based human simulation and gaze-guided vision-language pragmatics
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—
Output data is not yet available: Evolve's snapshot contains no full publication counts or yearly productivity metrics for this researcher beyond the two reviewed 2026 arXiv papers. This axis would measure total publication volume and annual output trends.
Impact—
Impact data is insufficient: no citation counts, h-index, or venue-level influence metrics are provided in this snapshot. This axis would measure citation-based influence of the researcher's work across the community; it cannot be scored from current evidence.
Focus0/100
Focus scores 0, the lowest reading on the radar, driven by the field distribution in the reviewed catalog: the two 2026 arXiv papers split evenly — 0.5 LLM-based human simulation and 0.5 vision-language pragmatics. With only two reviewed papers, this even split reads as diffuse rather than concentrated specialization; a deeper catalog would clarify whether the two threads are one integrated agenda.
Collaboration44/100
Collaboration scores 44 on a 0-100 scale, based on 8 distinct coauthors in the reviewed data. Both catalog papers are co-authored rather than solo, with Suhr in a middle-author position on one and last-author position on the other, neither corresponding. The frequent-coauthor roster — Téa Wright, Minwoo Kang, Seun Eisape, Ayush Raj, Suhong Moon, Joseph Suh, David M. Chan, and John Canny — points to a steady small-team working pattern.
Activity—
Activity data is not yet available: the catalog holds only two arXiv papers, both dated 2026, so no multi-year publishing cadence can be computed. This axis measures publication frequency and recent momentum; it remains unscored.
Reproducibility78/100
Reproducibility scores 78, computed as the mean repro-readiness across 2 repro-reviewed papers. The gaze-based referring paper rates high readiness, backed by strong ablations and human evaluation, while the persona-mixture human-simulation paper rates medium. Known gaps are unexamined data contamination in the latter and missing significance tests plus weak cross-backbone generalization in the former.
Research summary
Alane Suhr is a researcher at the University of California, Berkeley whose reviewed work sits at the intersection of two natural language processing branches: LLM-based human simulation and vision-language pragmatics. The Evolve catalog holds two of their 2026 arXiv papers, split evenly between these fields (0.5 each in the field distribution). One introduces pretrained Persona Mixture Models and tandem selection for human simulation: Evolve's review finds pretrained PMM base models outperform instruction-tuned LLMs, with tandem selection performing best, while flagging unexamined data contamination and medium reproducibility readiness. The other trains vision-language speakers for referring expression generation using estimated listener gaze scanpaths as reinforcement-learning rewards; strong ablations and human evaluation earn a high reproducibility rating, though significance tests are missing and cross-backbone generalization is weak. Radar scores report reproducibility at 78 (mean across 2 repro-reviewed papers), collaboration at 44 with 8 distinct coauthors, and focus at 0, reflecting the even 50/50 field split. Output, impact, and activity lack sufficient data. Frequent coauthors include Téa Wright, Minwoo Kang, Seun Eisape, Ayush Raj, Suhong Moon, Joseph Suh, David M. Chan, and John Canny.
Collaboration pattern
The reviewed data records 8 distinct coauthors and a collaboration score of 44 on a 0-100 scale. Both 2026 papers are co-authored, with Suhr in a middle-author position on the human-simulation paper and last-author position on the gaze paper, neither corresponding. The recurring coauthor roster includes Téa Wright, Minwoo Kang, Seun Eisape, Ayush Raj, Suhong Moon, Joseph Suh, David M. Chan, and John Canny, consistent with small-team work within UC Berkeley projects.
Research keywords
Highlights
Persona Mixture Models and Tandem Selection for Human Simulation
Learning to Refer from Estimated Listener Gaze
Timeline
- 2026Released 'Pretrained Persona Mixture Models and Tandem Models for Human Simulation' on arXiv, contributing as a middle author.
- 2026Released 'Learning to Refer from Estimated Listener Gaze' on arXiv, contributing as last author.
- 2026Both reviewed papers entered the Evolve catalog in the same year, spanning LLM-based human simulation and vision-language pragmatics in an even field split.
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.
- Bibliometric scores and raw metrics are computed from an OpenAlex-based snapshot; no other bibliometric sources are represented.
- Only two arXiv papers (both 2026) have been reviewed in the Evolve catalog, so output, impact, and activity axes lack sufficient data.
- No citation counts, h-index, or venue-level statistics are provided, so influence cannot be assessed.
- No ORCID or country field is available; the profile is name-resolved from public bibliographic data and identity disambiguation is limited.