Hao Zhang: ML researcher spanning multimodal VLM reasoning and process-based computer-use agent evaluation
2
papers here
2
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
—
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-08) 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—
Insufficient data: only two catalogued papers appear here, with no publication counts or venue record. This axis would measure cumulative productivity — total and per-year publication volume and venue quality — and can be scored once a fuller bibliographic record is linked.
Impact—
Insufficient data: no citation counts, h-index, or downstream-adoption signals are available here. This axis would measure how much later work builds on these papers — citations, adoption of OSWorld-Pro and SVGLM — and requires reliably matched citation records before scoring.
Focus0/100
The focus axis scores 0, and the field distribution explains why: the two catalogued papers split exactly evenly — 50% in 'AI / machine learning (agent evaluation and benchmarking)' and 50% in 'machine learning (multimodal vision-language models)'. With only two outputs, this reads as a deliberately two-sided profile — a benchmarking line (OSWorld-Pro) paired with a multimodal-reasoning line (Renderable Programs) — though whether breadth or concentration truly defines the agenda cannot yet be settled.
Collaboration62/100
The collaboration axis scores 62, grounded in 22 distinct coauthors across the reviewed record. The recurring pool spans both research lines — Zhilin Wang, Shaokun Zhang, Yifan Zhang, Jin Xu, Binfeng Xu, Jian Hu, and Yunheng Zou on one side; Karan Sapra, Andrew Tao, Jan Kautz, and Yi Dong on the other. On both catalogued papers Zhang sits in the middle of the author list rather than leading, so the network reflects participation in larger team-scale projects.
Activity—
Insufficient data: only 2026 appears in the record, so publication cadence over time cannot be characterized. This axis would measure research tempo — papers per year, year-over-year consistency, and recent momentum — and needs a multi-year publication history before scoring.
Reproducibility55/100
Reproducibility scores 55, computed as the mean readiness rating (reproMean 55) over the 2 reviewed papers, each rated 'medium'. Evolve's verdicts flag concrete gaps: the SVGLM paper reports ~6.5–10 weighted-point gains over SFT but states no significance tests and no release, while OSWorld-Pro validates its judge solidly yet relies on a single-judge, single-run design. Reproducers should treat both as promising but re-verify independently before building on the numbers.
Research summary
Hao Zhang is a machine learning researcher whose visible record centers on two complementary directions in modern AI systems: multimodal reasoning with vision-language models, and process-based evaluation of computer-use agents. Both papers in the Evolve catalog are 2026 arXiv preprints, and on each Zhang is a middle author without corresponding-author status. In 'Multimodal Thinking with Renderable Programs', the SVGLM line of work tunes VLMs to sketch SVG overlays while reasoning, reported to gain roughly 6.5–10 weighted points over supervised fine-tuning on geometry tasks — though Evolve notes that no significance tests or release are stated. In 'OSWorld-Pro', Zhang contributes to a process-based evaluation benchmark for computer-use agents spanning 305 tasks, 2,814 subgoals, and about 67,000 human labels, with solid judge validation but a single-judge, single-run design. Bibliometrically, the profile shows 22 distinct coauthors (collaboration score 62), an even 50/50 split across two fields (focus score 0), and a reproducibility score of 55 averaged over the two reviewed papers. Output, impact, and activity metrics are not yet computable from the available data.
Collaboration pattern
The record lists 22 distinct coauthors (collaboration score 62), and the recurring pool spans both research lines: Zhilin Wang, Shaokun Zhang, Yifan Zhang, Jin Xu, Binfeng Xu, Jian Hu, and Yunheng Zou on one side; Karan Sapra, Andrew Tao, Jan Kautz, and Yi Dong on the other. On both catalogued papers Zhang occupies a middle authorship slot without corresponding-author status, so the pattern reads as steady participation in larger team projects rather than solo-led work.
Research keywords
Highlights
SVG overlays as a reasoning medium for VLMs
Process-based evaluation for computer-use agents
Timeline
- 2026Middle-author contribution to 'Multimodal Thinking with Renderable Programs' (arXiv), applying SVG-overlay visual reasoning to vision-language models.
- 2026Middle-author contribution to 'OSWorld-Pro: Process-based Evaluation for Computer Use Agents' (arXiv), a benchmark of 305 tasks, 2,814 subgoals, and ~67k human labels.
- 2026Both catalogued outputs land in the same year, splitting evenly between multimodal VLM reasoning and agent evaluation/benchmarking.
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 are computed from public bibliometric data (OpenAlex-style); no citation counts, venue list, or h-index were provided.
- 'Hao Zhang' is a very common name; this profile is name-resolved from public bibliographic data and may include homonym records.
- No institution, country, or ORCID is attached, so identity and affiliation are unverified.
- Only 2 papers from the Evolve catalog underpin this profile; output, impact, and activity dimensions lack sufficient data and are not scored.