Tongji University researcher in LLM agents, efficient vision-language models, and robot learning (VLA).
4
papers here
3
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
0
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—
Insufficient data: OpenAlex output metrics are not yet usable for scoring, with h-index and i10-index both recorded as 0. This axis would measure total publication volume, venue quality, and productivity trends across the researcher's career.
Impact—
Insufficient data: citation impact metrics are not yet available — h-index and i10-index both read 0, indicating incomplete citation coverage. This axis would measure citation counts, h-index growth, and influence on subsequent research.
Focus0/100
Focus scores 0 because the OpenAlex field distribution splits evenly across three distinct areas — LLM-based software engineering agents, efficient vision-language models, and vision-language-action robot learning at 33.3% each. Topic data reinforces the breadth: topic modeling (15), multimodal machine learning applications (12), neural network applications (6), NLP techniques (6), and information retrieval and search behavior (5). The profile reflects deliberate multi-thread breadth rather than a single deep specialization.
Collaboration68/100
Collaboration scores 68. The OpenAlex record lists 29 distinct co-authors in the recent window, with 12 recurring collaborators including Haoran Que, Jiajun Shi, Jiaheng Liu, Ge Zhang, and Wenhao Huang. All four Evolve-catalogued 2026 papers place Wei Ye in a middle author position rather than first or corresponding, indicating work within larger team-based groups typical of large-model research.
Activity100/100
Activity scores 100, the strongest axis in this profile. OpenAlex records 33 works within the last three years, and the counts-by-year data places all 33 in 2026 within this snapshot. Four of these papers appear in the Evolve catalog as 2026 arXiv preprints spanning robotics, vision, and software engineering agents, confirming a dense and sustained submission cadence.
Reproducibility55/100
Reproducibility scores 55, averaged over 3 Evolve-reviewed papers, each rated medium reproducibility readiness. The pattern is consistent: GeoAAC ships no code release or significance tests, CoVeR's strong 8%-retention results come with weaker significance reporting, and AttnCompress's 53.17% pass rate at 33.6% lower cost versus AgentDiet remains significance-untested. Reported gains look coherent, but statistical rigor and artifact availability limit independent verification.
Research summary
Wei Ye is a researcher at Tongji University whose recent work concentrates on efficiency and inference-time adaptation in large-model systems. OpenAlex records 33 works in the last three years — all dated 2026 in this snapshot — produced with 29 distinct co-authors, and the field distribution splits evenly across three threads: LLM-based software engineering agents, efficient vision-language models, and vision-language-action robot learning. In the software engineering thread, AttnCompress introduces attention-guided rolling compression of agent trajectories, reporting 53.17% resolution at 33.6% lower cost than AgentDiet on SWE-Bench. In the vision thread, CoVeR applies coverage-based token pruning to multi-view 3D reasoning, with evidence strongest at 8% token retention. In the robotics thread, GeoAAC derives training-free adaptive action chunking for VLA policies from denoising-trajectory geometry. A fourth preprint, REER-PT, explores perplexity-guided pre-training data augmentation but has not yet been reviewed by Evolve. Evolve reviews describe gains as consistent but partly small, with significance testing and code release often missing; reproducibility averages 55 across three reviewed papers. Topic data also shows strength in topic modeling and multimodal machine learning applications, and all catalogued papers place Ye in a middle author position.
Collaboration pattern
Wei Ye works within sizeable teams: OpenAlex records 29 distinct co-authors in the recent window, with 12 recurring collaborators including Haoran Que, Jiajun Shi, Ting Huang, Renming Pang, Jiaheng Liu, Ge Zhang, and Wenhao Huang. Across all four catalogued 2026 arXiv papers, Ye appears as a middle author rather than first or corresponding, consistent with specialist contributions inside larger LLM, vision, and robotics groups rather than independently led papers.
Research keywords
Highlights
AttnCompress: attention-guided trajectory compression for software engineering agents
CoVeR: coverage-based token pruning for multi-view 3D reasoning
GeoAAC: geometry-based adaptive action chunking from denoising trajectories
Timeline
- 2026Released GeoAAC on arXiv, proposing training-free adaptive action chunking for VLA policies from denoising-trajectory geometry.
- 2026Released CoVeR on arXiv, introducing coverage-based token pruning for multi-view 3D reasoning in vision-language models.
- 2026Released AttnCompress on arXiv, applying attention-guided trajectory compression to autonomous software engineering agents.
- 2026Preprinted REER-PT on perplexity-guided pre-training data augmentation via reverse-engineered reasoning; not yet reviewed by Evolve.
Papers in the Evolve catalog
GeoAAC: Geometry-Based Adaptive Action Chunking from Denoising Trajectories in VLA Policies
CoVeR: Coverage-Based Token Pruning for Multi-View 3D Reasoning in VLMs
AttnCompress: Dynamic Attention-Guided Trajectory Compression for Software Engineering Agents
REER-PT: Reverse-Engineered Reasoning for Perplexity-Guided Pre-training Data Augmentation
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.
- Bibliometrics come from OpenAlex only; h-index and i10-index are recorded as 0, which likely reflects incomplete citation coverage rather than a verified absence of impact.
- All 33 recent works are dated 2026 in this snapshot, so longer-term productivity trends cannot be assessed.
- "Wei Ye" is a relatively common name; this profile is name-resolved from public bibliographic data (OpenAlex identity, ORCID 0000-0002-9331-4716).
- REER-PT has not been reviewed by Evolve; its verdict, keywords, and reproducibility readiness are unassessed, and it has no catalog slug.