LLM-agents researcher: teaching models when to ask clarifying questions and how to estimate GUI-action confidence
2
papers here
2
reviewed by Evolve
0
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 scored: this profile lacks the underlying bibliometric record (paper counts, venues, years of output) needed to measure publication volume. The axis would track how much and how steadily this researcher publishes, which cannot be computed from the current two-paper catalog.
Impact—
Not scored: no citation counts, h-index, or downstream-adoption data are available in the input. The axis would measure the influence of this researcher's work on later research, which cannot be estimated from a two-paper arXiv record alone.
Focus0/100
The focus score is 0, driven by an even 50/50 field distribution: one cataloged paper sits in machine learning (GUI agents, uncertainty estimation) and the other in natural language processing (LLM agents, reinforcement learning). With only two papers, this split reads as breadth across two adjacent agent-oriented topics rather than deep concentration in one niche; more papers are needed to tell whether the interests are converging.
Collaboration22/100
Collaboration scores 22/100, grounded in a very small co-author network: the record lists 2 distinct coauthors, with Yunxiang Li and Xixin Wu recurring across the catalog. Both papers place Helen Meng in the last-author slot, though neither credits a corresponding author. The pattern suggests a compact, stable core team rather than wide external collaboration; the low score reflects network size, which a two-paper sample will understate.
Activity—
Not scored: the input contains no longitudinal publication timeline spanning multiple years, so recent publishing cadence and continuity cannot be measured. The axis would track whether this researcher is currently active, which requires a longer run of dated records than the catalog provides.
Reproducibility55/100
Reproducibility is the strongest scored dimension at 55/100, averaged over 2 reviewed papers, both rated medium readiness by Evolve. The reviews temper this: CIGAsk's gains are solid but single-seed and simulator-dependent, and PACE's improvements are real yet modest, with 2 of 6 bootstrap confidence intervals crossing zero. The axis measures how far methods can be re-implemented and results reproduced; both papers sit mid-range rather than high.
Research summary
This profile is name-resolved from public bibliographic data: Helen Meng's institution, country, and ORCID are unresolved, and no biographical facts are asserted beyond the cataloged record. Evolve's catalog holds two 2026 arXiv papers, each listing Helen Meng as last author and neither marked corresponding. The work splits evenly — a 50/50 field distribution — between machine learning (GUI agents, uncertainty estimation) and natural language processing (LLM agents, reinforcement learning). CIGAsk trains LLM agents to decide when and how to ask clarifying questions, combining counterfactual information gain against frozen references with an ambiguity bonus inside GRPO; Evolve's review calls the gains solid but flags single-seed training and simulator dependence. PACE weights digit-level entropy by place value to estimate GUI click confidence; its gains are real but modest — 2 of 6 bootstrap confidence intervals cross zero — and the method is scoped to coordinate formats. Scores mirror the thin record: collaboration 22/100 from only 2 distinct coauthors (Yunxiang Li and Xixin Wu recurring), focus 0 from the even two-field split, and reproducibility 55/100 averaged over both medium-readiness papers. Output, impact, and activity are unscored for lack of data. Read this as an early-stage record of a compact, tightly collaborative LLM-agents research line, not a complete career picture.
Collaboration pattern
The record shows a small, stable footprint: 2 distinct coauthors, with Yunxiang Li and Xixin Wu recurring across both papers. Helen Meng sits in the last-author position on each, without being the corresponding author — a placement conventionally associated with a senior role, though the data does not confirm one. The collaboration score of 22/100 reflects this narrow two-person network; nothing in the record evidences wider institutional or consortium collaboration.
Research keywords
Highlights
CIGAsk: teaching agents when and how to clarify
PACE: place-aware coordinate entropy for click confidence
Timeline
- 2026CIGAsk posted to arXiv: GRPO-trained clarifying behavior for LLM agents, with Helen Meng as last author.
- 2026PACE posted to arXiv: place-aware coordinate entropy for GUI-agent click confidence, again with Helen Meng as last author.
- 2026Both cataloged papers land in the same year, and Evolve rates each at medium reproducibility readiness.
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 scores are computed from OpenAlex-derived bibliographic records only; no citation counts, venues, or publication-volume metrics were available.
- Identity is name-resolved from public bibliographic data; institution, country, and ORCID are unresolved, so this record should not be assumed to describe any particular known researcher sharing the name.
- The evidence base is 2 arXiv papers from 2026 reviewed by Evolve; output, impact, and activity dimensions remain unscored for lack of data.
- Reproducibility and quality judgments reflect Evolve's own paper reviews, not independent citation or replication data.