University of Macau researcher in LLM alignment: efficient fine-tuning, multilingual reward models, agentic memory
3
papers here
2
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—
Not yet scored: full output metrics (lifetime publication counts by year and venue type) are unavailable in the current OpenAlex record beyond 20 works in the recent three-year window. This axis would measure overall publication volume and consistency.
Impact—
Not yet scored: OpenAlex reports an h-index of 0 and i10-index of 0, so citation coverage is too thin to quantify influence. This axis would measure citation impact, uptake, and scholarly influence of the work.
Focus0/100
Scores 0 on this dataset's concentration measure: the field distribution splits evenly (0.5/0.5) between LLM fine-tuning and reward modeling, and OpenAlex topics spread across five clusters — Topic Modeling (13), NLP Techniques (8), AI Problem Solving and Planning (7), Multimodal ML (5), and Domain Adaptation/Few-Shot Learning (4) — indicating deliberately broad, early-stage exploration rather than a single niche.
Collaboration62/100
Scores 62: the OpenAlex record lists 22 distinct coauthors, and 12 collaborators recur across projects, including Weihua Luo, Junchao Wu, Yichao Du, Derek F. Wong, and Jinsong Su. In the three reviewed papers Longyue Wang appears as a middle-position, non-corresponding author, consistent with contribution inside larger team-led LLM research groups.
Activity89/100
Scores 89, the strongest axis in the record: OpenAlex logs 20 works in the recent three-year window, with the counts-by-year series concentrated in 2026 (20 works). All three Evolve-reviewed papers appeared on arXiv in 2026, spanning fine-tuning, reward modeling, and agentic memory — a high-tempo, current output pattern.
Reproducibility55/100
Scores 55, based on Evolve's reproducibility review of 2 papers (mean 55), both rated medium readiness. Recurring verdict notes include absent significance testing (NGFT, UniRRM), no variance tests (CHIME), and LLM-verified data circularity risk (UniRRM) — method code and results are plausible to rerun, but statistical rigor gaps remain.
Research summary
Longyue Wang is a researcher at the University of Macau working on large language model alignment and LLM-based agents. The OpenAlex record (name-resolved from public bibliographic data; no linked ORCID) shows 20 works in the recent three-year window, all logged in 2026, alongside 22 distinct coauthors — a high-activity, team-based early-stage profile. Within Evolve's catalog, three 2026 arXiv papers span the alignment stack: Neuron-Guided Fine-Tuning (NGFT) uses neuron activations to unify instruction data selection, sparse updates, and anti-forgetting in supervised fine-tuning; UniRRM, with the MixReward dataset of 64.5K preferences across 103 languages, unifies pairwise, listwise, and point-wise reward judging across languages and evaluation paradigms; and CHIME introduces credit-aware hierarchical memory evolution for long-horizon agentic planning, reporting 2.96–3.68 point gains over self-evolving baselines. Evolve's reproducibility review of two papers averages 55/100 with medium readiness, with recurring notes on missing significance or variance testing and potential data circularity in LLM-verified preference construction. Topic distribution (Topic Modeling 13, NLP Techniques 8, AI Problem Solving and Planning 7) confirms breadth across NLP and agents. Citation-based indicators (h-index 0, i10-index 0) are too thin to score output or impact, so this profile should be read as an early-stage snapshot of an actively publishing alignment researcher.
Collaboration pattern
The OpenAlex record lists 22 distinct coauthors, with 12 recurring collaborators including Weihua Luo, Junchao Wu, Yichao Du, Derek F. Wong, Runzhe Zhan, Shu Yang, and Jinsong Su. All three Evolve-reviewed papers place Longyue Wang as a middle-position, non-corresponding author, suggesting a contributor role inside larger, senior-led LLM research teams rather than solo-led projects.
Research keywords
Highlights
Neuron-guided fine-tuning unifies the alignment pipeline
Multilingual reward judging at 103-language scale
Credit-aware hierarchical memory for long-horizon agents
Timeline
- 2026arXiv preprint on Neuron-Guided Fine-Tuning (NGFT) proposes neuron activations as a unified lever for data selection, sparse updates, and anti-forgetting in SFT.
- 2026UniRRM and the MixReward dataset (64.5K preferences, 103 languages) unify pairwise, listwise, and point-wise reward judging across languages and evaluation paradigms.
- 2026CHIME introduces credit-aware hierarchical memory evolution for long-horizon agentic planning, reporting 2.96–3.68 point gains over self-evolving baselines.
- 2026OpenAlex logs 20 works in 2026 with 22 distinct coauthors, marking the most active year in the record.
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.
- Bibliometric indicators are derived from OpenAlex only; no ORCID is linked, so identity is name-resolved from public bibliographic data.
- OpenAlex reports h-index 0 and i10-index 0, so citation coverage is too thin to score output or impact; both axes are marked missing.
- All three reviewed papers are 2026 arXiv preprints; peer-reviewed venue status is not reflected in the input data.
- The profile draws on Evolve's reviews of 3 papers; coverage of the researcher's full publication record is not guaranteed.