CAS researcher on efficient MoE inference and 3DGS/GPU efficiency; early-stage, strong reproducibility record
2
papers here
2
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
0
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-16) 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—
Publication-volume metrics are not yet available in this snapshot, so output is unscored. This axis would measure total works, annual production trajectory, and authorship roles; the catalog currently covers only two 2026 arXiv preprints.
Impact—
Impact cannot be scored: the OpenAlex record lists h-index and i10-index as 0, consistent with very recent, not-yet-cited preprints. This axis would measure citation counts, citing works, and influence on follow-up research once citations accrue.
Focus0/100
The focus axis scores 0 because output is evenly split: field distribution is 50% computer systems (computer architecture and graphics) and 50% machine learning (efficient LLM inference), with three OpenAlex topics (parallel computing, graphics and visualization, 3D shape modeling) each appearing once. At this early stage the record shows breadth across two efficiency-oriented areas rather than a single dominant specialization, so no concentration signal can be measured yet.
Collaboration63/100
The collaboration axis scores 63 from 23 distinct coauthors in the snapshot. Twelve recurring partners appear in the catalog record, including Peisong Wang, Qiwei Dong, Minnan Pei, and Yihan Zhou, indicating a stable core team behind both 2026 preprints; both papers carry the researcher in the last author position, a pattern consistent with senior or supervisory roles within a moderately sized CAS-affiliated group.
Activity20/100
The activity axis scores 20 on thin evidence: counts-by-year shows a single work in 2026 and recent-3-year works equal 1, so both catalog papers fall in the same year with no longer indexed history. This likely reflects a newly separated OpenAlex identity or recent entry into indexed publishing rather than a true productivity pause; the baseline will firm up as later works accrue.
Reproducibility78/100
Reproducibility is this profile's strongest scored axis at 78, the mean across 2 papers assessed by Evolve. From Splats to Silicon rates high reproducibility readiness, while the MoE expert-pruning study rates medium; both are arXiv preprints whose empirical claims rest on disclosed settings, giving an unusually transparent early record even though independent replications are not yet documented.
Research summary
Jian Cheng is a researcher affiliated with the Chinese Academy of Sciences whose indexed record, resolved from OpenAlex, currently spans two 2026 arXiv preprints on which they serve as last author. The two papers define a dual-track efficiency agenda. In machine learning, an empirical study of dynamic expert pruning in fine-grained mixture-of-experts LLMs reports that uniformly retaining 2/3 of the chosen experts preserves 98.8% of MoE performance, with dynamic pruning paying off only under aggressive budgets — an iso-budget evaluation that Evolve judged convincing and solid. In computer systems, 'From Splats to Silicon' offers a workload-centric re-examination of the computational efficiency of 3D Gaussian splatting, combining a survey with original GPU profiling across rasterization and sorting stages; Evolve called it rigorous and unusually honest while noting a narrow empirical base of one GPU and four scenes. Field distribution is evenly split, 50% computer systems (architecture and graphics) and 50% machine learning (efficient LLM inference). The record shows 23 distinct coauthors with a recurring core of twelve, and a reproducibility mean of 78 across two assessed papers (one high, one medium readiness). Citation metrics are not yet meaningful — h-index and i10-index both stand at 0 — so this is best read as an early-stage profile whose quality signals come from assessed content rather than accumulated impact.
Collaboration pattern
Jian Cheng publishes as last author on both catalog papers, alongside 23 distinct coauthors overall. The recurring roster includes Peisong Wang, Minnan Pei, Qiwei Dong, Yihan Zhou, Gang Li, Yuchen Zhu, Wenju Zhao, Zhongtian Long, Siting Wang, Yuanteng Chen, Qiwei Lai, and Chen Tianqi. The two papers sit in different fields (GPU/graphics systems and efficient LLM inference), suggesting either two collaborating sub-teams or a co-supervision setup rather than one fixed lab line.
Research keywords
Highlights
Uniform 2/3 expert retention keeps 98.8% of MoE performance
A workload-centric efficiency account of 3D Gaussian splatting
One senior-author seat across a dual-track efficiency agenda
Both 2026 catalog papers place Jian Cheng in the last author position across two different fields — efficient LLM inference and computer architecture/graphics — supported by a recurring pool of twelve coauthors, a pattern consistent with a coordinating or supervisory role in a two-direction efficiency effort.
Timeline
- 2026Last-author arXiv preprint reporting an empirical study of dynamic expert pruning in fine-grained MoE LLMs
- 2026Last-author arXiv preprint 'From Splats to Silicon' rethinking the computational efficiency of 3D Gaussian splatting
- 2026OpenAlex counts-by-year records 1 indexed work with 23 distinct coauthors, and the reproducibility mean across two assessed papers stands at 78
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 an OpenAlex identity record only; 'Jian Cheng' is a common Chinese name, so this profile is name-resolved from public bibliographic data and could merge or split homonymous authors.
- Coverage is thin: only two 2026 arXiv preprints are in the snapshot, so output and impact are unscored and h-index and i10-index both stand at 0.
- The reproducibility mean of 78 is computed over just 2 assessed papers, and the 3DGS survey's own empirical base is one GPU and four scenes.
- No ORCID is linked, and the peer-review or venue status of both arXiv preprints is unknown.