Huawei (China) researcher advancing self-evolving LLM agents, auditable agent memory, and web-agent environments
4
papers here
0
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
1
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—
Output cannot be scored yet: the bibliometric record is thin (h-index 1, i10-index 0, countsByYear covering only 2024 and 2026). This axis would measure total publication volume, venue quality, and productivity trajectory as fuller indexing accumulates.
Impact—
Impact is not scored: with an h-index of 1 and an i10-index of 0, citation evidence is insufficient. This axis would measure citation counts, citing works, and influence on follow-up research; none of these signals has accumulated yet.
Focus—
Focus is not scored: OpenAlex topics (AI-based Problem Solving and Planning 5; Topic Modeling, NLP, Information Retrieval 4 each) span several areas, too young for a stable core. This axis would measure how consistently output concentrates in one primary specialization.
Collaboration70/100
OpenAlex records 33 distinct coauthors for Yang Liu, and the radar collaboration dimension scores 70. The frequent-coauthor list (Wen Jiang, Mingmin Chu, Yimeng Tian, Qianxin Zhang, Haofei Yang, Rui Yang, Tao Lv, Fangming Li, Kejia Zhang, Tianyuan Zou, Zixuan GU, Junxuan Li) suggests a broad Huawei-centered team pattern. All four cataloged 2026 papers place Yang Liu in middle or last positions, none corresponding, indicating participation within larger group efforts rather than solo leadership so far.
Activity83/100
The activity dimension scores 83, the strongest signal in this profile. OpenAlex counts 16 works in the last three years, with countsByYear jumping from 2 works in 2024 to 14 in 2026 — a steep acceleration. All four Evolve-cataloged papers are 2026 arXiv releases, consistent with a fast, preprint-first cadence typical of an early-career industrial researcher ramping up.
Reproducibility—
Reproducibility is not scored. The only signal available is Evolve's rating of low reproducibility readiness for the HarnessEvolve paper; the other three cataloged papers lack assessments. This axis would measure code, data, and benchmark artifact availability across the body of work.
Research summary
Yang Liu is a researcher at Huawei Technologies (China) working on large language model agents. This profile is name-resolved from OpenAlex bibliographic data; no ORCID was available. OpenAlex records 16 works in the last three years, 14 of them in 2026, up from 2 works in 2024 — a steep acceleration consistent with an early-career burst, reflected in an activity score of 83. The citation footprint remains minimal (h-index 1, i10-index 0), so impact is not yet scored. The Evolve catalog includes four 2026 arXiv papers: HarnessEvolve, which evolves the agent harness from reference trajectories with dual gates and is reported to beat GEPA, ACE, and SkillOpt on 5 benchmarks without significance tests; Hindsight Memory-PRM, supervising memory management with auditable hindsight credit; Learning Simple Test-Time Environments for LLM Web Agents; and Auditing and Mitigating Privacy Leakage in Cloud-Edge Collaborative Decoding. Authorship sits in middle or last positions, none corresponding. Topics span AI-based problem solving and planning (5), topic modeling, NLP techniques, and information retrieval (4 each), plus reinforcement learning in robotics (3), with 33 distinct coauthors and a collaboration score of 70.
Collaboration pattern
Yang Liu works in a broad, team-based industrial setting: OpenAlex lists 33 distinct coauthors and the collaboration dimension scores 70. Twelve frequent coauthors — including Wen Jiang, Mingmin Chu, Yimeng Tian, Qianxin Zhang, Haofei Yang, and Rui Yang — anchor a Huawei-centered network. Across all four cataloged 2026 papers Yang Liu appears in middle or last author positions with no corresponding authorship, suggesting intensive collaboration within group projects rather than independently led efforts so far.
Research keywords
Highlights
Reference-trajectory self-evolution for agent harnesses
Auditable hindsight credit for agent memory management
A 2026 arXiv paper proposing to supervise memory management with auditable hindsight credit (Hindsight Memory-PRM); Evolve has not yet published a verdict, keywords, or reproducibility assessment for it.
Simple test-time environments for LLM web agents
A 2026 arXiv paper with Yang Liu as last author proposing to learn simple test-time environments for LLM web agents; no Evolve verdict or reproducibility assessment is available yet.
Privacy leakage in cloud-edge collaborative decoding
A 2026 arXiv paper with Yang Liu as last author that audits and mitigates privacy leakage in cloud-edge collaborative decoding; no Evolve verdict or reproducibility assessment is available yet.
Timeline
- 2024First recorded works appear in the OpenAlex countsByYear record, with 2 publications in 2024.
- 2026Output accelerates sharply: OpenAlex records 14 works in 2026, and the activity dimension reaches 83.
- 2026Releases HarnessEvolve on arXiv — reference-trajectory-based agent self-evolution with dual gates, in a middle author position.
- 2026Publishes Hindsight Memory-PRM plus two further arXiv papers as last author: simple test-time environments for LLM web agents, and auditing privacy leakage in cloud-edge collaborative decoding.
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 metrics are computed from OpenAlex only (identity: openalex); no ORCID was available for disambiguation.
- "Yang Liu" is a very common name; this profile is name-resolved from public bibliographic data and may mix or omit same-name records.
- countsByYear coverage is sparse (2024: 2 works; 2026: 14 works), so career-stage and trajectory interpretations are provisional.
- Only four 2026 arXiv papers have been reviewed in the Evolve catalog; three of them lack verdicts, keywords, and reproducibility assessments.