RL researcher focused on credit assignment and entropy-based hallucination mitigation for LLMs and multimodal models
2
papers here
0
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
0
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-09-28) 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 measures publication volume and productivity over time. This profile cannot score it: OpenAlex reports an h-index and i10-index of 0 with only two indexed works, a footprint too sparse to characterize Weiming Liu's research output.
Impact—
Impact measures citation uptake, citations per year, and influence on subsequent work. No citation counts are available in this snapshot; with an h-index and i10-index of 0, the axis cannot be scored yet.
Focus—
Focus measures how concentrated research activity is across topics and venues. Four OpenAlex topic labels exist, covering topic modeling, multimodal ML applications, RL in robotics, and advanced GNNs, but with only two indexed works the sample is too thin to score.
Collaboration50/100
OpenAlex records 11 distinct coauthors for Weiming Liu, and the Evolve catalog lists 11 frequent collaborators including Xincheng Yao, Haobo Fu, and Chongyang Zhang. Both cataloged 2026 papers place Liu in a middle author position rather than as corresponding author, consistent with a team-based contributor pattern; the raw coauthor breadth supports the moderate collaboration score of 50.
Activity32/100
Activity is scored at 32. OpenAlex shows two works in the recent three-year window, with year-by-year counts containing only 2026 (2 works), meaning both cataloged papers appeared in the same year and no earlier or later output is indexed. This narrow, single-year footprint sits below typical peer baselines and drives the below-median activity score.
Reproducibility—
Reproducibility would measure code, data, and seed availability across outputs. The profile-level signal is insufficient to score; per-paper Evolve reviews rate both works as medium reproducibility readiness, with DEEPO specifically lacking a code release.
Research summary
Weiming Liu is a machine-learning researcher whose recent work centers on reinforcement learning for large language models, with an emphasis on credit assignment and hallucination control. The Evolve catalog contains two 2026 arXiv papers with Liu in a middle author position, neither as corresponding author. The first, GRAFT, reformulates step-level advantage estimation for agentic RL by merging rollouts into a trajectory graph whose Bellman iterations produce step-level advantages; Evolve's review judges the theory and three-seed experiments promising but thin, with medium reproducibility readiness. The second, DEEPO, targets hallucination in multimodal LLMs through dual-entropy enhanced policy optimization built on GRPO, combining semantic entropy with Renyi entropy; the review finds the evidence unusually rigorous but flags cross-table number inconsistencies and no code release. OpenAlex bibliometrics show an h-index and i10-index of 0, two works in the recent three-year window (both 2026), and eleven distinct coauthors, consistent with an early-stage, team-based research profile. Topic labels, including topic modeling, multimodal ML applications, reinforcement learning in robotics, and advanced graph neural networks, hint at breadth beyond the two cataloged papers, though the indexed sample is too small to characterize a research focus.
Collaboration pattern
Weiming Liu's recorded collaboration footprint is broad relative to output: 11 distinct coauthors in OpenAlex and 11 frequent collaborators, including Xincheng Yao, Haobo Fu, Chongyang Zhang, Yingxuan Zhuang, Miao Pan, Wangjie Gan, Jingxiao Yang, Fan Wang, Cheng Tan, Xuhong Zhang, and Jintao Chen. On both cataloged 2026 papers Liu sits in a middle author position and is not the corresponding author, indicating contribution within larger GRPO-focused teams rather than solo leadership.
Research keywords
Highlights
GRAFT: trajectory graphs for step-level advantages in agentic RL
DEEPO: dual-entropy policy optimization against MLLM hallucination
A recurring GRPO credit-assignment thread across both papers
Both 2026 cataloged papers build on GRPO and attack the credit-assignment problem, one via graph-structured advantage estimation for agentic RL and the other via entropy-based signals for hallucination correction, suggesting a coherent methodological agenda.
Timeline
- 2026GRAFT appears on arXiv, proposing trajectory-graph Bellman estimation of step-level advantages for agentic reinforcement learning.
- 2026DEEPO appears on arXiv, introducing dual-entropy enhanced policy optimization to mitigate hallucination in multimodal LLMs.
- 2026Evolve reviews both papers; the OpenAlex snapshot records 2 works in the recent three-year window, 11 distinct coauthors, and an h-index and i10-index of 0.
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.
- Identity is name-resolved from public OpenAlex bibliographic data; no ORCID or institutional affiliation is available, so attribution for a common name carries uncertainty.
- Bibliometrics come from OpenAlex only; the output and impact axes are unscored because indexed works and citations are too sparse.
- The catalog covers two 2026 arXiv preprints reviewed by Evolve; no peer-reviewed venue record appears in this snapshot.
- Focus and reproducibility radar scores are omitted rather than estimated; topic labels cover only four OpenAlex categories from a minimal sample.