Researcher in scientific machine learning, efficient LLM architectures, and computational attention modeling
3
papers here
2
reviewed by Evolve
2
first-author
1
corresponding
—
works · OpenAlex
0
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-17) 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: aggregate output metrics are unavailable yet. This axis measures total publication volume, year-by-year distribution, venue mix, and authorship roles. The reported h-index of 0 contradicts 43 works in 2026, so no estimate is made.
Impact—
Insufficient data: citation-based impact cannot yet be scored. This axis tracks total citations, h-index, and field-normalized influence over time. Only h-index and i10-index of 0 are reported with no citation counts by year, so no estimate is made.
Focus0/100
OpenAlex splits the field distribution exactly 50/50 between scientific machine learning / computational mechanics of metamaterials and efficient LLM architectures, and the catalog adds a third strand in visual cognition — yet the raw topic vector is dominated by particle-physics labels (counts 32, 32, 21, 9) inconsistent with the ML papers. That incoherence drives the focus score to 0/100, more likely reflecting record mixture for this common name than genuine scattering.
Collaboration84/100
The score (84/100) rests on 66 distinct co-authors in the recent window alongside 43 recent works, so nearly every output brings new collaborators. Cataloged papers span a large consortium credit (DeepSeek-AI, where this author is a middle author) and first-authored academic teams whose recurring partners include Alexander Ogren, Cynthia Rudin, Johann Guilleminot, and L. Catherine Brinson. The pattern is high-throughput, team-based, and crosses institutional boundaries.
Activity100/100
The maximum score (100/100) reflects raw publication cadence: OpenAlex counts 43 works in 2026 alone, equal to the recent-3-year total of 43, so essentially all recorded output concentrates in the current year. All three Evolve-reviewed papers appeared on arXiv in 2026, two of them first-authored. Such a single-year spike signals intense current activity, but it also means longer-term productivity trends cannot yet be assessed.
Reproducibility60/100
The score (60/100) is the mean over two assessed papers (reproMean 60, reproPapers 2). In the Evolve catalog, the wavelet-encoded Fourier neural operator paper and the goal-directed attention paper rate high reproducibility readiness, while the DeepSeek-V4.1-Flash KV cache paper rates low — self-evaluated with no ablations. Recurring reviewer concerns are single-run experiments and missing external baselines, which temper otherwise strong practices.
Research summary
Han Zhang (ORCID 0000-0002-7752-8538), affiliated in this data with the Michigan Department of Transportation in the US, is profiled from public bibliographic records and three 2026 arXiv papers reviewed by Evolve. OpenAlex logs 43 works in 2026 and 66 distinct co-authors, with the field distribution split 50/50 between scientific machine learning / computational mechanics of metamaterials and efficient large language model architectures. The reviewed output spans three strands: a first-authored paper using wavelet-encoded Fourier neural operators to predict metamaterial eigenmodes with a claimed ~1000x speedup (high reproducibility readiness); a first- and corresponding-authored study reporting that zero-shot vision-language models align with human goal-directed attention across 4,887 scenes (high readiness); and a middle-author contribution to DeepSeek-V4.1-Flash KV cache compression (low readiness). Evolve's reproducibility mean over two assessed papers is 60/100. Radar scores show maximal activity (100/100) and strong collaboration (84/100), while focus is 0/100 because OpenAlex attaches particle-physics topic labels to a record whose reviewed papers are ML-focused — a mismatch suggesting identity mixture for this very common name. Output and impact are unscored: the reported h-index of 0 conflicts with the 2026 output volume, so citation coverage appears incomplete.
Collaboration pattern
With 66 distinct co-authors recorded against 43 recent works, Han Zhang operates in large, fast-rotating teams. Recurring named partners include Alexander Ogren, Cynthia Rudin, Johann Guilleminot, and L. Catherine Brinson, alongside a DeepSeek-AI consortium credit. The catalog pairs first-authored academic preprints with a middle-author industrial collaboration. Some listed co-author strings (e.g., ':') are parsing artifacts, and name mixture may inflate the count, so treat 66 as an upper bound.
Research keywords
Highlights
Metamaterial eigenmodes learned with wavelet-encoded Fourier neural operators
Zero-shot vision-language models align with human goal-directed attention
KV cache compression at frontier scale in DeepSeek-V4.1-Flash
Timeline
- 2026First-authored arXiv preprint: wavelet-encoded Fourier neural operators learn metamaterial eigenmodes, with strong ablations and a claimed ~1000x speedup.
- 2026First and corresponding author on Emergent Goal-Directed Attention in Large Vision-Language Models, evaluated across 4,887 scenes.
- 2026Middle-author contribution to DeepSeek-V4.1-Flash KV cache compression, a claimed 552B-parameter MoE system.
- 2026OpenAlex logs 43 works this year with 66 distinct co-authors — all recorded output concentrated in a single year.
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 from OpenAlex only; the reported h-index of 0 alongside 43 works in 2026 indicates incomplete citation coverage.
- 'Han Zhang' is a very common name; this profile is name-resolved from public bibliographic data, and the particle-physics topic labels conflict with the reviewed ML papers, indicating likely record mixture.
- All three catalog papers are 2026 arXiv preprints; peer-review status is not reflected in this data.
- The co-author list contains parsing artifacts (e.g., ':' and consortium entries), so the 66 distinct co-author figure should be read as an upper bound.