Kuaishou researcher spanning LLM agents, multimodal GUI agents, and inference serving for generative recommendation.
3
papers here
3
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
0
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-16) 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 metrics are not yet available: the snapshot provides no reliable publication-count series beyond three 2026 arXiv records. This axis would measure cumulative works, per-year productivity, and output trajectory across venues.
Impact—
Impact data is insufficient: the snapshot reports zeroed citation aggregates (h-index and i10-index both 0), which more likely reflects incomplete author-name resolution than genuine lack of influence. This axis would measure citations, downstream uptake, and field influence.
Focus0/100
The focus score of 0 reflects maximal topical spread in the underlying data: field attribution divides evenly at one-third across LLM-based agents, multimodal GUI agents, and ML inference serving, and OpenAlex topic assignments span five distinct areas (topic modeling, parallel computing, recommender systems, RL in robotics, usability) with one record each. The 2026 cataloged papers cohere more tightly around agents and serving, but the recorded footprint is broad rather than concentrated.
Collaboration80/100
Collaboration scores 80, the profile's strongest axis: OpenAlex records 55 distinct coauthors, and the catalog lists 12 recurring collaborators, including Longtao Hu, Linchao Zhu, and Tong Ye. All three 2026 papers position Liang as a middle author rather than corresponding, consistent with contribution inside sizeable industry teams at Kuaishou rather than solo or small-group work.
Activity40/100
Activity scores 40: the snapshot records 3 works in the last three years, all concentrated in 2026 — the counts-by-year series contains a single entry (2026: 3). Within that year output is dense, with three arXiv preprints, but the visible series is too short to confirm sustained multi-year research activity.
Reproducibility32/100
Reproducibility scores 32, the mean across 3 assessed papers. Evolve rates OneLA and BlueLM-GUI as low readiness — the former evaluated on a single model, the latter carrying self-referential benchmarking and a confounded ablation — while the OSWorld skill-evolution paper reaches medium, aided by a control condition despite single-run confounds. Real-device and industry-scale setups currently limit independent replication.
Research summary
Xiao Liang is a researcher at Kuaishou (China) whose cataloged work sits at the intersection of large language model agents and machine learning systems. All three papers in Evolve's catalog are 2026 arXiv preprints, and OpenAlex field attribution splits evenly three ways: LLM-based agents, multimodal GUI agents, and inference serving, each at one third. The portfolio covers OneLA, which scales linear-attention (Gated DeltaNet) decoding to large beams in generative recommendation with reported 1.54–2.46× decode speedups; BlueLM-GUI, a real-device flywheel for self-improving mobile GUI agents reporting 87.4 on its own benchmark and 84.9 on AndroidWorld; and an online skill-evolution study for computer-use agents claiming +5.7–18.6 percentage points over an empty-library control on OSWorld. On all three, Liang is a middle-position author and not corresponding, consistent with a contributor role inside larger industry teams. OpenAlex records 3 works in the last three years, all in 2026, alongside 55 distinct coauthors, a collaboration score of 80, and a reproducibility mean of 32 across the 3 assessed papers. Output and impact metrics are not yet reliably available.
Collaboration pattern
Xiao Liang works within sizeable industry teams: OpenAlex records 55 distinct coauthors, and the catalog lists 12 recurring collaborators, among them Longtao Hu, Linchao Zhu, Tong Ye, Kunyang Han, and Guozhi Wang. On all three 2026 papers Liang is a middle author and not corresponding, suggesting a specialist contributor role within Kuaishou's agent and serving groups rather than leading authorship. The collaboration dimension scores 80, the strongest axis in this profile.
Research keywords
Highlights
OneLA: linear-attention decoding at scale for generative recommendation
BlueLM-GUI: real-device flywheel for self-improving mobile GUI agents
Evolving persistent skills for computer-use agents on OSWorld
Timeline
- 2026Co-authored OneLA, scaling linear-attention decoding to large beams in generative recommendation (arXiv).
- 2026Co-authored the BlueLM-GUI technical report on real-device, self-improving mobile GUI agents (arXiv).
- 2026Co-authored work on online skill evolution from interaction traces for computer-use agents (arXiv).
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.
- Name resolution: 'Xiao Liang' is a common name; this profile is name-resolved from public bibliographic data (OpenAlex) and may merge or split distinct researchers.
- Bibliometrics come from OpenAlex only; the reported h-index and i10-index are 0, likely an artifact of incomplete author disambiguation rather than measured impact.
- Paper assessments reflect Evolve's reviews of three 2026 arXiv preprints; no peer-reviewed venue information is present in this snapshot.
- Country is not recorded in the source data; the institution is listed as Kuaishou (China).