Bowen Zhou

Shanghai Artificial Intelligence Laboratory · Australia

Unclaimed · AI-structured profileOpenAlex

Shanghai AI Laboratory researcher in efficient long-context LLM inference and verifiable agent environments

papers here

reviewed by Evolve

first-author

corresponding

—

works · OpenAlex

h-index · OpenAlex

Bibliometric figures come from OpenAlex (2026-10-15) 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 data is insufficient to score: the OpenAlex record shows an h-index of 0 with no cumulative publication counts, so productivity cannot be estimated. This axis would measure total works, venue quality, and yearly output distribution once fuller bibliographic metadata is linked.

Impact—

Impact cannot be scored: the record's citation fields are effectively empty, with an h-index of 0 and i10-index of 0, reflecting missing citation linkage rather than measured influence. This axis would measure citations, citation velocity, and referencing works across the paper portfolio.

Focus0/100

The focus dimension scores 0, reflecting wide topical dispersion in the OpenAlex record: five topic clusters led by Multimodal Machine Learning Applications (14 works), Topic Modeling (9), Advanced Neural Network Applications (7), GANs and Image Synthesis (5), and NLP Techniques (4). The two Evolve-reviewed papers likewise split the field distribution evenly at 0.5/0.5 between LLM agents and sparse-attention inference, indicating parallel research tracks rather than one narrow line.

Collaboration75/100

Scoring 75, collaboration rests on 42 distinct coauthors in the recent window and a recurring core of eleven named collaborators, including Shuai Bai, Jiayong Deng, Sicheng Fan, Yikun Fu, Chang Gao, and Mianqiu Huang. Both Evolve-reviewed papers place Bowen Zhou in the last author position on large arXiv team efforts, a signature consistent with supervisory or group-lead roles in sizeable laboratory teams rather than solo work.

Activity89/100

The activity score of 89 is grounded in 20 works over the recent three-year window, with the OpenAlex counts-by-year field recording all 20 in 2026 alone. Both Evolve-reviewed arXiv preprints appeared in 2026 — one on efficient sparse attention, one on agent environments — indicating a dense, current output cadence concentrated in the most recent year of the record.

Reproducibility78/100

Reproducibility averages 78 across the two Evolve-reviewed papers (reproPapers: 2). RecreationWorld rates high readiness, with Evolve judging its five-platform evaluation design rigorous even though transfer gains lack statistical controls; the block-sparse attention paper rates medium, with strong reported speed (5.03x at 128K context) but thin gains over Prism and unreported variance. The pattern suggests generally reusable artifacts with statistical-reporting gaps in the inference work.

Research summary

Bowen Zhou is a researcher affiliated with the Shanghai Artificial Intelligence Laboratory whose recent work spans two complementary tracks in large language model systems: efficient long-context inference and environments for computer-use agents. The OpenAlex record attributes 20 works over the recent three-year window, all dated 2026, and distributes topics across five clusters led by Multimodal Machine Learning Applications (14), Topic Modeling (9), and Advanced Neural Network Applications (7), while the two Evolve-reviewed papers split evenly (0.5/0.5) between LLM agents and sparse-attention inference. Both reviewed 2026 arXiv preprints place Zhou in the last author position. RecreationWorld introduces a five-platform benchmark plus SFT pipeline for hybrid GUI/coding agents with verifiable rewards, rated high reproducibility readiness; the block-sparse attention paper proposes training-free pre-RoPE block routing with offline geometric priors, reporting 5.03x speedup at 128K context, rated medium readiness. Reproducibility averages 78 across the two reviewed papers. Collaboration is broad: 42 distinct coauthors and eleven recurring collaborators, with a collaboration score of 75 and an activity score of 89. Output and impact are not yet scorable because the record's citation fields (h-index and i10-index of 0) lack linkage; this profile is name-resolved from OpenAlex bibliographic data.

Collaboration pattern

Bowen Zhou's collaboration footprint is broad: 42 distinct coauthors in the recent window and a recurring core of eleven named collaborators, including Shuai Bai, Jiayong Deng, Sicheng Fan, Yikun Fu, Chang Gao, and Mianqiu Huang. Both Evolve-reviewed 2026 papers place Zhou in the last author position on large arXiv teams, consistent with a supervisory role, and the collaboration dimension scores 75.

Research keywords

block-sparse attentiontraining-free routinglong-context LLM inferencerotary position embedding (RoPE)phase interferenceoffline calibrationcomputer-use agentsGUI agentshybrid coding agentsverifiable rewardsagent benchmark environments

Highlights

Training-free block-sparse attention reaches 5.03x speedup at 128K context
The SGDR paper routes pre-RoPE attention blocks with offline geometric priors, decoupling semantic and geometric signals for long-context inference. Evolve rates the speed gains strong at 128K but finds the margin over Prism thin and variance unreported, assigning medium reproducibility readiness.
RecreationWorld: a verifiable five-platform environment for hybrid computer-use agents
This 2026 paper builds a five-platform recreation benchmark plus an SFT pipeline for hybrid GUI/coding agents with verifiable rewards. Evolve judges the evaluation design rigorous and rates reproducibility readiness high, while noting that transfer gains lack statistical controls.
Dual-track 2026 output across inference efficiency and agent environments
The record shows 20 works in 2026 with 42 distinct coauthors, and the field distribution splits evenly (0.5/0.5) between LLM agents and sparse-attention inference. Both reviewed papers list Bowen Zhou as last author, pairing a systems-speed track with an agent-benchmark track in the same year.

Timeline

2026Released RecreationWorld on arXiv, a five-platform verifiable environment and SFT pipeline for hybrid GUI/coding agents, in the last author position.
2026Published block-sparse attention with semantic-geometric decoupled routing on arXiv, reporting a 5.03x speedup at 128K context.
2026OpenAlex counts-by-year records 20 works in 2026 with 42 distinct coauthors, the densest and only fully dated year in the available record.

Papers in the Evolve catalog

Frequent co-authors

Shuai Bai×1 Jiayong Deng×1 Sicheng Fan×1 Yikun Fu×1 Chang Gao×1 Mianqiu Huang×1 Mianqiu Huang×1 Yizhen Jiang×1 Yuheng Jing×1 Dehui Kong×1 Keliang Li×1 Ning Li×1

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 derive from an OpenAlex record only; no ORCID is linked, and Bowen Zhou is a common name, so identity is name-resolved from public bibliographic data.
Citation fields show an h-index of 0 and i10-index of 0 with year counts only for 2026, so the output and impact dimensions are left unscored rather than estimated.
The record's country field reads AU while the institution is listed as Shanghai Artificial Intelligence Laboratory; this affiliation-country pairing is unverified.
Only two papers carry Evolve review verdicts; topic counts and the field distribution reflect OpenAlex assignment, not a curated publication list.