Jintao Chen

Zhejiang University · China

Unclaimed · AI-structured profileOpenAlex

Zhejiang University researcher optimizing LLM-agent policies with reinforcement learning and credit assignment.

papers here

reviewed by Evolve

first-author

corresponding

—

works · OpenAlex

h-index · OpenAlex

Bibliometric figures come from OpenAlex (2026-10-17) 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: the snapshot reports an h-index and i10-index of 0 and no year-by-year volume beyond 2026. This axis would measure total works, productivity trend, and venue spread once a longer indexed record exists.

Impact—

Impact is unscored because citation-derived signals — an h-index and i10-index of 0 — offer no usable distribution here. This axis would capture citation accumulation, uptake in follow-up work, and field-level recognition once the 2026 papers accrue citations.

Focus100/100

Focus scores 100, the maximum, reflecting a profile concentrated in a single declared field — machine learning with reinforcement learning for LLM agents. Recent topic counts show topic modeling (4), reinforcement learning in robotics (3), and AI-based planning (3) leading, yet both reviewed 2026 papers sit squarely in agent-RL, indicating one coherent current research line.

Collaboration51/100

Collaboration scores 51, a mid-range value. The snapshot lists 12 distinct coauthors in the recent window, with recurring partners including Yingxuan Zhuang, Jingxiao Yang, Wenqi Zhang, Xuhong Zhang, and Jianwei Yin. On both reviewed papers Chen is a non-corresponding co-author — last on BATON, middle on TASPO — consistent with contributing inside a larger Zhejiang University group.

Activity72/100

Activity scores 72. The countsByYear series records 11 works in 2026, matching the recent3yWorks figure of 11, and both catalog entries are 2026 arXiv preprints released close together. This points to a concentrated burst of current output rather than a demonstrated long-term publication rhythm, which the single-year snapshot cannot confirm either way.

Reproducibility55/100

Reproducibility scores 55, computed from one paper in Evolve's reproducibility sample (reproPapers: 1, mean 55). Both reviewed preprints carry a medium repro-readiness rating: BATON's statistics and theory are described as thin, while TASPO's results are strong but draft artifacts lower trust. This remains a small evidence base for a reliability score.

Research summary

Jintao Chen is a machine-learning researcher at Zhejiang University whose recent work centers on reinforcement learning for large language model (LLM) agents. An OpenAlex-resolved snapshot records 11 works in the trailing three-year window, all attributed to 2026, with topics led by topic modeling (4), reinforcement learning in robotics (3), and AI-based problem solving and planning (3), alongside natural language processing (2) and GAN-based image synthesis (2). Evolve has reviewed two 2026 arXiv preprints from this record. BATON reweights token losses through Bayesian feedback attribution and equalizes trajectory mass, delivering consistent GRPO/GiGPO gains, though its statistics and theory remain thin. TASPO reconciles process supervision with outcome-based credit and reports a 10.6% average improvement across nine agent-RL settings, with strong ablations offset by draft artifacts. Chen is a non-corresponding co-author on both — last position on BATON, middle on TASPO — suggesting contribution within a larger team. The profile shows maximal focus (100) and solid activity (72), while output and impact remain unscored because h-index and i10-index have not yet accumulated.

Collaboration pattern

Chen's collaboration footprint spans 12 distinct coauthors in the recent OpenAlex window, anchored at Zhejiang University. The most frequent partners — Yingxuan Zhuang, Jingxiao Yang, Xuhong Zhang, Wenqi Zhang, Jianwei Yin, Wangjie Gan, Binhe Yu, Ruopei Sun, Ziting Li, and Cheng Tan — form a dense local cluster around LLM-agent reinforcement learning. Both reviewed papers are multi-author efforts in which Chen is not the corresponding author (last position on BATON, middle on TASPO), suggesting a contributor role within a team rather than independent group leadership.

Research keywords

LLM agentsreinforcement learningpolicy optimizationcredit assignmentGRPOGiGPOprocess supervisionagentic policy optimizationon-policy self-distillationprivileged information

Highlights

BATON: dual-axis policy optimization for LLM agents
Evolve's review finds that BATON reweights token losses via feedback-likelihood evidence and equalizes trajectory mass across rollouts, yielding consistent gains over GRPO and GiGPO baselines; its statistics and theoretical analysis remain thin, and reproducibility readiness is rated medium.
TASPO: reconciling process supervision with outcome-based credit
TASPO injects privileged action-level supervision into GRPO credit assignment, reporting a 10.6% average improvement across nine agent-RL settings with strong ablations; Evolve notes that draft artifacts lower trust, with medium reproducibility readiness.
A concentrated 2026 line on credit assignment for agent RL
Both reviewed 2026 preprints attack the same problem — credit assignment in reinforcement learning for LLM agents — from complementary angles: trajectory-level loss shaping in BATON versus privileged process supervision in TASPO, situating Jintao Chen within a focused Zhejiang University research effort.

Timeline

2026OpenAlex records 11 works in the trailing three-year window, all attributed to 2026.
2026TASPO preprint appears on arXiv, reporting a 10.6% average gain across nine agent-RL settings with Chen in a middle author position.
2026BATON preprint appears on arXiv, introducing dual-axis policy optimization via Bayesian feedback attribution and trajectory mass normalization, with Chen in the last author position.

Papers in the Evolve catalog

Frequent co-authors

Yingxuan Zhuang×2 Jingxiao Yang×1 Wangjie Gan×1 Wenqi Zhang×1 Xuhong Zhang×1 Binhe Yu×1 Jingxiao Yang×1 Ruopei Sun×1 Ziting Li×1 Cheng Tan×1 Xuhong Zhang×1 Jianwei Yin×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.

All bibliometrics derive from a single OpenAlex snapshot; no ORCID or cross-database validation was available.
The countsByYear series contains only 2026 (11 works), so long-term productivity and citation history fall outside this snapshot's scope.
Output and impact are unscored because the recorded h-index and i10-index are both 0 — an absence of citation signal, not negative evidence.
"Jintao Chen" is a common Chinese name; this profile is name-resolved from public OpenAlex bibliographic data plus the two Evolve-reviewed papers.