UIUC researcher in multimodal vision-language reasoning and LLM evidence fusion for computational social science
2
papers here
2
reviewed by Evolve
0
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—
Not yet scored: this profile has no reliable counts of total publications, venues, or output volume, so the output axis cannot be computed. It would measure cumulative publication rate and productivity across the researcher's full bibliographic record.
Impact—
Not yet scored: no citation counts, h-index, or venue-level influence data are available for this name-resolved profile. The impact axis would measure citations, downstream adoption, and scholarly attention to the researcher's published work.
Focus0/100
Scores 0 on focus: the Evolve record splits evenly, 50% natural language processing / computational social science and 50% machine learning (multimodal vision-language models), across the two cataloged 2026 papers. With only two papers, one in each area, this even split reads as breadth rather than settled specialization; a larger corpus would be needed to tell whether Liu is genuinely cross-field or simply early in consolidating a direction.
Collaboration63/100
Scores 63 on collaboration, based on 23 distinct coauthors across the cataloged work, with 12 recurring collaborators including Yuhe Wu, Rui Qian, Guangyu Wang and others. On both 2026 papers Liu sits in the middle of the author list and is not the corresponding author, a pattern consistent with contributing as a team member inside larger groups rather than leading independent projects.
Activity—
Not yet scored: yearly publication volumes and recency trends are not available beyond the two 2026 catalog papers. The activity axis would measure publishing cadence and recent output intensity across the researcher's career timeline.
Reproducibility55/100
Scores 55, the mean readiness across 2 reproducibility-scored papers, both rated medium. The main caveat is transparency rather than method complexity: the SVGLM paper states no significance tests or code release, and TEF's gains rest on weak baselines and synthetic data. The methods themselves — entropy-weighted log-odds fusion and SVG-overlay fine-tuning — are described simply enough that released artifacts could lift this score.
Research summary
Jiaxin Liu is a researcher at the University of Illinois Urbana-Champaign whose Evolve-reviewed work to date spans two adjacent areas of machine learning: multimodal reasoning with vision-language models, and natural language processing applied to computational social science. The current catalog contains two 2026 arXiv papers, in both of which Liu is a middle-listed, non-corresponding author, consistent with a contributor role on larger teams. The first, 'Multimodal Thinking with Renderable Programs', introduces SVGLM, which tunes vision-language models to sketch SVG overlays while reasoning, reported to gain roughly 6.5-10 weighted points over SFT on geometry tasks. The second, 'Count Evidence, Not Sentences', proposes Tempered Evidence Fusion (TEF), an entropy-weighted aggregation of sentence-level log-odds from LLM judgments that beats naive voting by about 4.5 accuracy points on long-text value measurement. Evolve rates both papers medium on reproducibility readiness, yielding a mean of 55 across 2 scored papers, with neither paper stating significance tests or code release. Bibliometrically, the profile shows 23 distinct coauthors and an even 50/50 split between the two fields, producing a low focus concentration score.
Collaboration pattern
Liu works in sizeable teams: 23 distinct coauthors across the two cataloged papers, with a recurring core of 12 collaborators (Yuhe Wu, Rui Qian, Guangyu Wang, Yuran Chen, Yuanchao Zhu, Junjie Yang, among others). Sitting in the middle author position on both 2026 papers, without corresponding authorship, suggests a specialist contributor role — likely owning experiments or modeling components — within groups whose projects span both multimodal vision-language work and NLP for computational social science.
Research keywords
Highlights
SVGLM: renderable SVG overlays as a thinking medium for vision-language models
Tempered Evidence Fusion: entropy-weighted LLM judgments for long-text value measurement
A dual-track 2026: structured visual reasoning and calibrated evidence aggregation
Across the two 2026 catalog papers, Liu contributes to a shared methodological instinct: converting messy signals — images on one side, sentence-level LLM judgments on the other — into structured, aggregatable representations. Both papers are middle-authored, arXiv-sourced, and rated medium on reproducibility readiness.
Timeline
- 2026Cataloged paper 'Multimodal Thinking with Renderable Programs' (SVGLM) appears on arXiv, with Liu as a middle author in multimodal vision-language research.
- 2026Cataloged paper 'Count Evidence, Not Sentences' (TEF) appears on arXiv, with Liu as a middle author in NLP for computational social science.
- 2026The cataloged record closes the year split 50/50 across two fields, with 23 distinct coauthors and a reproducibility mean of 55 across 2 scored papers; output, impact and activity metrics remain unscored for lack of data.
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 come from a sparse OpenAlex-derived snapshot: only 2 cataloged papers, with no citation, h-index, or venue-level data.
- 'Jiaxin Liu' is a common name; this profile is name-resolved from public bibliographic data and lacks ORCID disambiguation.
- Both cataloged papers are 2026 arXiv preprints with Liu as middle, non-corresponding author; contribution claims are inferred from authorship position only.
- Output, impact and activity radar axes are unscored due to insufficient data; collaboration (63), focus (0) and reproducibility (55) are Evolve's computed scores from the raw metrics shown.