LLM-agents researcher at CUHK: teaching models when to clarify and estimating GUI-agent click confidence
2
papers here
2
reviewed by Evolve
2
first-author
0
corresponding
—
works · OpenAlex
—
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-13) 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: the catalog holds only two 2026 arXiv preprints, with no publication counts, venue record, or volume metrics available. This axis would measure total research output — papers, venues, and productivity over time.
Impact—
No citation counts, h-index, or downstream-usage signals exist in the data, so impact cannot be scored. This axis would measure how much the two clarification-learning and GUI-confidence papers are cited, adopted, or built upon by others.
Focus0/100
The focus axis scores 0, while the field distribution records an even 50/50 split between machine learning (GUI agents, uncertainty estimation) and natural language processing (LLM agents, reinforcement learning). With only two papers, this zero reflects breadth across two adjacent lines rather than demonstrated scattering: both works target LLM agents, one on when to clarify and one on click confidence. The score should be read as an early-stage signal, not a settled pattern.
Collaboration22/100
The collaboration score is 22, based on two distinct coauthors in the raw data — Xixin Wu and Helen Meng — and a first-author (non-corresponding) position on both cataloged papers. The footprint is small and concentrated, consistent with a researcher embedded in one CUHK group; the axis measures breadth of the co-author network, not quality of ties.
Activity—
Activity is unscored: with only two 2026 arXiv preprints and no multi-year publication history, there is too little signal on publishing cadence. This axis would measure output rhythm over time — preprints, submissions, and updates per year.
Reproducibility55/100
Evolve records a mean reproducibility score of 55 across two papers (reproPapers=2), both rated medium readiness. CIGAsk is flagged as single-seed and simulator-dependent; PACE shows two of six bootstrap confidence intervals crossing zero, with results scoped to coordinate formats. The mid-range score reflects real but partially fragile empirical support.
Research summary
Yunxiang Li is a researcher at the Chinese University of Hong Kong whose cataloged work centers on large language model (LLM) agents, spanning two adjacent lines recorded at 0.5 each in the field distribution: natural language processing (LLM agents, reinforcement learning) and machine learning (GUI agents, uncertainty estimation). Both papers in the Evolve catalog are first-authored arXiv preprints from 2026. The first, CIGAsk, teaches LLM agents when and how to ask clarifying questions, combining frozen-reference information gain with an ambiguity bonus inside GRPO training; Evolve notes solid gains but flags single-seed results and simulator dependence. The second, PACE, introduces place-aware coordinate entropy for GUI agent confidence estimation, weighting digit-level entropy by place value; Evolve judges the gains real but modest, with two of six bootstrap confidence intervals crossing zero and scope limited to coordinate formats. Across the two papers Evolve records a mean reproducibility score of 55 (n=2), with both rated medium reproducibility readiness. The recorded collaboration footprint is narrow: two distinct coauthors, Xixin Wu and Helen Meng, with first-author positions on both papers. Output, impact, and activity axes remain unscored for insufficient data. This is an early-stage profile built from a small bibliometric snapshot.
Collaboration pattern
The recorded collaboration footprint is small: two distinct coauthors, Xixin Wu and Helen Meng, across both cataloged papers, with Yunxiang Li in the first-author (non-corresponding) seat on each. This single-group pattern, anchored at the Chinese University of Hong Kong, fits an early-career researcher working closely within one lab rather than an extended external network. The collaboration dimension score of 22 reflects this narrow breadth; external collaborations may emerge as the catalog grows.
Research keywords
Highlights
CIGAsk: teaching LLM agents when and how to ask clarifying questions
PACE: place-aware coordinate entropy for GUI agent confidence estimation
Timeline
- 2026Publishes CIGAsk on arXiv, a first-author study teaching LLM agents when and how to clarify via counterfactual information gain.
- 2026Publishes PACE on arXiv, a first-author study on place-aware coordinate entropy for GUI agent confidence estimation.
- 2026Both papers enter the Evolve catalog with medium reproducibility readiness and a mean reproducibility score of 55.
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.
- Bibliometrics derive from a small snapshot: two 2026 arXiv preprints reviewed by Evolve; no citation, venue, or career-history data are available.
- Output, impact, and activity dimensions are unscored because the underlying data is insufficient.
- ORCID is not provided and 'Yunxiang Li' is a common name; this profile is name-resolved from public bibliographic data and institution affiliation only.
- Collaboration data cover only the Evolve-cataloged papers; co-author activity outside the catalog is not represented.