Yiming Wang — efficient large-model inference systems, calibrated expression recognition, and robotic manipulation
3
papers here
1
reviewed by Evolve
1
first-author
0
corresponding
—
works · OpenAlex
—
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-09-21) 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—
No output score is available yet. This axis would measure total publication volume and venue quality across the career; the current Evolve snapshot holds only three 2026 arXiv records, too few to compute, so no estimate is made.
Impact—
Impact is not yet scored. It would measure citations, citation velocity, and influence on follow-up work; none of the three cataloged 2026 papers carries citation data in this snapshot, so the axis is left unscored.
Focus100/100
Focus scores 100/100. The bibliometric field distribution places the entire labeled record in a single field, ML systems / efficient large-model inference (count: 1), and the reviewed paper's keywords — mixture-of-experts, SSD expert offloading, routing prediction, int4 quantization, recovery LoRA distillation, on-device inference — reinforce one systems-centered agenda. Two further 2026 papers touch vision and robotics, so the perfect score rests on a thin labeled base and a broader portfolio.
Collaboration63/100
Collaboration scores 63/100, with 23 distinct coauthors in the snapshot. Recurring names include Frederick W. B. Li, Jingyun Wang, Lijie Wang, Zheng Lu, Heyang Yu, Kenghou Hoi, Bowen Hu, Di Cui, Tianyu Xin, and Haoran Liao; 'Yiming Wang' also appears twice as a coauthor, hinting at a namesake. Wang takes first-author position on one 2026 paper but middle positions on the reviewed systems work, consistent with team-based research.
Activity—
No activity score is available yet. This axis would measure recent publication cadence and ongoing research throughput; all three records fall within 2026, giving too narrow a time window to judge sustained productivity, so it is left unscored.
Reproducibility100/100
Reproducibility scores 100/100, computed as a mean over one reviewed paper (reproPapers: 1). That paper — the SSD-resident MoE serving work — carries a 'high' reproducibility readiness from Evolve's review, which praised unusually honest reporting of what the system does and does not beat. With a single paper behind the score, treat it as an early, encouraging signal rather than an established track record.
Research summary
Yiming Wang is an early-career researcher whose Evolve catalog record spans ML systems for efficient large-model inference, computer vision, and robotic manipulation. The most complete record is a 2026 arXiv paper, 'The Other Half of the Memory Wall: Serving 35B MoEs from SSD with Trained Routing Prediction', where Wang is a middle-position author. Evolve's review credits the Edge0 system with serving a 35B MoE from SSD at 20 tok/s inside 3 GiB via one-token-ahead routing prediction plus an unmerged recovery LoRA, and calls it solid, unusually honest systems work while noting the headline gains lack external offloading baselines; its reproducibility readiness is rated high. Wang also first-authored a 2026 preprint on predicting human disagreement for calibrated dynamic facial expression recognition, and co-authored 'Acting in Meters: Learning Metric Interactions for Precise Robotic Manipulation'. Bibliometric signals show 23 distinct coauthors (collaboration 63/100), a focus score of 100 concentrated on the single labeled field — ML systems / efficient large-model inference — and a reproducibility score of 100 drawn from one reviewed paper. Output, impact, and activity scores are not yet computed, and no institution or country is recorded.
Collaboration pattern
Wang's snapshot lists 23 distinct coauthors (collaboration 63/100), with recurring partners including Frederick W. B. Li, Jingyun Wang, Lijie Wang, Zheng Lu, Heyang Yu, Kenghou Hoi, Bowen Hu, Di Cui, Tianyu Xin, and Haoran Liao. Wang first-authors the expression-recognition preprint and holds middle-author positions on the systems and robotics papers, suggesting a contributor within small research teams. Note that 'Yiming Wang' appears twice among the coauthors, so some entries may be a namesake.
Research keywords
Highlights
Serving a 35B MoE from SSD inside 3 GiB (Edge0)
Predicting human disagreement for calibrated facial expression recognition
A 2026 arXiv preprint with Wang as first author. The title indicates work on modeling annotator disagreement to calibrate dynamic facial expression recognition; Evolve has not yet reviewed it, so no performance claims are made here.
Metric interactions for precise robotic manipulation
Co-authored 2026 arXiv paper 'Acting in Meters: Learning Metric Interactions for Precise Robotic Manipulation'; the title points to learning metric-scale interaction representations for manipulation precision. It is not yet reviewed by Evolve.
Timeline
- 2026First-authored arXiv preprint 'Predicting Human Disagreement for Calibrated Dynamic Facial Expression Recognition'.
- 2026Co-authored the Edge0 paper serving 35B MoEs from SSD with trained routing prediction; Evolve rates its reproducibility readiness high.
- 2026Co-authored 'Acting in Meters: Learning Metric Interactions for Precise Robotic Manipulation'.
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.
- Radar scores and raw counts derive from a limited OpenAlex-based bibliometric snapshot; output, impact, and activity are not yet computed.
- 'Yiming Wang' is a common name and this profile is name-resolved from public bibliographic data; 'Yiming Wang' also appears twice among the coauthors, so some records may belong to a namesake.
- Only one of the three cataloged papers carries a full Evolve review (verdict, keywords, reproducibility readiness); the other two are title-level records without slugs.
- Institution, country, and ORCID are not recorded, and all cataloged papers are 2026 arXiv preprints — no peer-reviewed venue data is available yet.