Pengkun Jiao

Alibaba Group (China) · China

Unclaimed · AI-structured profileOpenAlex

Alibaba researcher spanning agentic RL for LLM agents and training-free sparse attention for long-context inference

papers here

reviewed by Evolve

first-author

corresponding

—

works · OpenAlex

h-index · OpenAlex

Bibliometric figures come from OpenAlex (2026-10-09) 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-output metrics are not yet available for scoring. This axis would measure the volume, venue quality, and consistency of published work; the OpenAlex record indexed under this name is too thin to support a reliable output score.

Impact—

Citation-impact data is insufficient for scoring. This axis would measure citations, citation trajectory, and scholarly influence; the only related fields on file are an h-index of 1 and an i10-index of 1, too sparse to score.

Focus0/100

Focus scores 0, and the field distribution explains why: work splits exactly 50/50 between reinforcement learning for LLM agents and efficient long-context LLM inference via sparse attention. Underlying OpenAlex topics add further spread — multimodal machine learning applications (11), advanced neural network applications (9), topic modeling (5), domain adaptation and few-shot learning (4), and adversarial robustness (3). The portfolio reads as broad, application-driven work rather than a single concentrated line.

Collaboration58/100

The collaboration score of 58 rests on 17 distinct coauthors in the OpenAlex record. The frequent-coauthor list is headed by Yiran Zhong, Feida Zhu, and Steven Hoi, with Feida Zhu and Steven Hoi each appearing more than once — a repeated-partner pattern typical of team-based industrial research. Both catalog papers list the researcher in a middle author position, consistent with contributing inside a larger Alibaba team rather than solo leadership.

Activity75/100

Activity scores 75 on the strength of recent output. OpenAlex counts 12 works over the past three years, with yearly counts of 3 (2024), 3 (2025), and 6 (2026) — a doubling in the most recent year. Both Evolve-catalog entries are 2026 arXiv preprints, matching this acceleration. The cadence suggests an active production phase, though volume alone says little about venue acceptance.

Reproducibility55/100

Reproducibility averages 55 across the 2 papers scored. Both carry a medium repro-readiness rating: ProCredit shows consistent three-scale AppWorld gains and clean ablations but is bounded by verifier scope, while SGDR reports a 5.03x speedup at 128K context yet shows thin gains over Prism with variance unreported. Code or data availability is not confirmed in either verdict.

Research summary

Pengkun Jiao is a researcher at Alibaba Group (China) whose indexed work, per OpenAlex, splits evenly between two directions: reinforcement learning for LLM agents and efficient long-context LLM inference via sparse attention. The bibliometric record shows 12 works over the past three years, with yearly counts rising from 3 (2024) and 3 (2025) to 6 (2026) — an acceleration consistent with an early-stage industrial research profile. Two 2026 arXiv preprints anchor the Evolve-reviewed catalog: ProCredit converts verifiable acceptance checks into per-turn progress rewards for agentic reinforcement learning, with consistent gains across three AppWorld scales and clean ablations; SGDR is a training-free block-sparse attention routing scheme reporting a 5.03x speedup at 128K context, though gains over Prism are thin and variance unreported. Topic frequencies cluster around multimodal machine learning applications (11), advanced neural network applications (9), and topic modeling (5), reflecting breadth rather than concentration, and the field distribution is exactly 50/50 across the two named areas. Collaboration is moderate (17 distinct coauthors), anchored by repeated partners such as Feida Zhu and Steven Hoi. Citation-based output and impact are not yet scorable: the indexed h-index and i10-index are both 1, so this profile should be read as early-stage.

Collaboration pattern

The OpenAlex record lists 17 distinct coauthors. The frequent-coauthor list is led by Yiran Zhong, Feida Zhu, and Steven Hoi, with Feida Zhu and Steven Hoi each appearing twice, indicating stable repeat partnerships. Both catalog papers place the researcher in a middle author position without corresponding status, consistent with contributing within a larger Alibaba research team rather than leading independently.

Research keywords

agentic reinforcement learningcredit assignmentverifiable rewardsreward shapingGRPOLLM agentsblock-sparse attentiontraining-free routinglong-context inferenceRoPE (rotary position embedding)multimodal machine learningadversarial robustness

Highlights

ProCredit turns verifiable checks into per-turn progress rewards
In agentic reinforcement learning for LLM agents, ProCredit converts verifiable acceptance checks into per-turn progress credit rather than a single outcome reward, with consistent gains across three AppWorld scales and clean ablations; scope remains bounded by the verifier.
Training-free block-sparse attention routing delivers 5.03x at 128K
SGDR routes pre-RoPE attention blocks with offline geometric priors, requiring no training, and reports a 5.03x speedup at 128K context; Evolve notes the gains over Prism are thin and variance is unreported.
Accelerating 2026 output across two LLM research tracks
OpenAlex records 6 works in 2026 — double the 3 of each prior year — including the two arXiv preprints reviewed here, spanning agentic RL reward design and efficient long-context inference.

Timeline

20243 works indexed in OpenAlex — the earliest year in the provided counts-by-year data.
20253 works indexed, matching the prior year's volume.
20266 works indexed — a doubling of yearly output — including both Evolve-reviewed arXiv preprints: ProCredit on agentic RL credit assignment and SGDR on block-sparse attention.

Papers in the Evolve catalog

Frequent co-authors

Yiran Zhong×2 Feida Zhu×2 Steven Hoi×2 Ming Ma×1 Yi Zhu×1 Chonghan Liu×1 Qichao Wang×1 Yanhao Jia×1 Tianming Yang×1 Feida Zhu×1 Pengkun Jiao×1 Steven Hoi×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 are drawn from OpenAlex only; identity is name-resolved from public bibliographic data (no ORCID on file), so records may mix same-name authors.
Output and impact dimensions are unscored: citation data on file is too sparse (h-index 1, i10-index 1) to characterize influence.
Only the 2 catalog papers carry Evolve verdicts and repro-readiness ratings; other indexed works are unreviewed.
Affiliation and country are taken as provided (Alibaba Group, China) and are not independently verified.