Early-footprint Xiaomi researcher across efficient LLM inference architecture and RL-based coding agents
2
papers here
2
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
1
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-13) 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—
Output cannot be scored yet: OpenAlex attribution shows an h-index of 1 and 8 works recorded in 2026, too sparse for a reliable productivity estimate. This axis would measure total and per-year publication volume once author-paper disambiguation improves.
Impact—
Impact cannot be scored yet: the resolved OpenAlex record shows an i10-index of 0 and h-index of 1, so citation coverage for this identity is absent. This axis would measure citation counts, h-index trajectory, and downstream influence.
Focus0/100
The focus axis scores 0, reflecting an evenly split field distribution: 50% efficient large language model inference (architecture design) and 50% machine learning (LLM coding agents). OpenAlex topics likewise spread across five areas — Natural Language Processing Techniques (5), Topic Modeling (4), Advanced Neural Network Applications (4), plus one work each in parallel computing and cloud resource management — indicating two parallel research tracks rather than a single narrow specialization.
Collaboration71/100
The collaboration axis scores 71 out of 100. The OpenAlex record links this author to 34 distinct coauthors, with a recurring core of 12 frequent collaborators including Hailin Zhang, Liang Zhao, Jianyu Wei, Yizhao Gao, and Yu Cheng. Both reviewed 2026 papers carry a last author position without corresponding-author status, a pattern consistent with contribution inside a large industrial research group.
Activity64/100
The activity axis scores 64. The OpenAlex record shows 8 works within the recent three-year window, with the per-year counts placing all of them in 2026, indicating a concentrated recent burst rather than a sustained multi-year cadence. Both cataloged papers also appeared in 2026, so measured activity reflects a very recent, front-loaded publishing pattern whose longer-term trajectory is not yet observable.
Reproducibility20/100
The reproducibility axis scores 20, the mean readiness across 2 reviewed papers, each rated low. Evolve flags an MQA-switch confound in HySparse2's comparisons and single-run results without significance testing in CodeMidas, both of which hinder independent verification. The score rests on a two-paper sample of brand-new arXiv preprints, so it is an early signal rather than a stable measure.
Research summary
Fuli Luo is a researcher at Xiaomi (China) working at the intersection of efficient large language model inference and reinforcement learning for coding agents. The OpenAlex-resolved record splits the field distribution evenly between these tracks — 50% efficient LLM inference (architecture design) and 50% machine learning (LLM coding agents) — and the evidence base is recent and small: 8 works recorded in 2026, with two arXiv preprints from that year reviewed in the Evolve catalog. HySparse2 proposes hybrid sparse attention with two-level KV cache sharing, cutting prefill cost and KV cache footprint while improving retrieval, though the MQA switch confounds comparisons; CodeMidas claims source code alone can yield 5,545 reinforcement learning environments that lift MiMo-V2.5 on five benchmarks, though results are single-run without significance tests. Collaboration is the strongest measured dimension at 71/100, with 34 distinct coauthors and a recurring core of 12 frequent collaborators; both reviewed papers carry last authorship inside what appears to be a large industrial team. Focus scores 0 because output splits evenly across the two fields, and reproducibility scores 20 with both papers rated low readiness. Citation-backed output and impact cannot be scored — the resolved record shows an h-index of 1 and i10-index of 0 — making this an early-stage profile whose footprint is not yet fully captured.
Collaboration pattern
This author works in large industrial teams, appearing in the last author position on both reviewed 2026 papers without corresponding-author flags. The OpenAlex record links 34 distinct coauthors, with a stable core of 12 frequent collaborators — including Hailin Zhang, Liang Zhao, Jianyu Wei, Yizhao Gao, Yu Cheng, and Yifan Song — spanning the inference-systems and RL-for-coding efforts.
Research keywords
Highlights
HySparse2: two-level KV sharing for long-context sparse attention
CodeMidas: 5,545 RL environments mined from code itself
Timeline
- 2026OpenAlex records 8 works for this author within the recent three-year window, with all cataloged activity dated to this year.
- 2026CodeMidas appears on arXiv, claiming 5,545 RL environments synthesized from source code alone that lift MiMo-V2.5 on five benchmarks.
- 2026HySparse2 appears on arXiv, introducing hybrid sparse attention with two-level KV sharing that cuts prefill and KV cache while improving retrieval.
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 indicators come from OpenAlex only; no ORCID is linked, so identity is resolved by name from public bibliographic data.
- Citation-backed metrics are sparse (h-index 1, i10-index 0), so the output and impact axes are left unscored rather than rated low.
- Both reviewed papers are 2026 arXiv preprints with low reproducibility readiness, so the reproducibility score rests on only two papers.
- Recorded topics such as Topic Modeling (4) do not appear among the reviewed papers, suggesting possible name-disambiguation noise in the OpenAlex record.