Huawen Shen

Tencent (China)

Unclaimed · AI-structured profileOpenAlex

Tencent researcher on vision-language GUI agents: on-policy self-distillation and masked trajectory prediction

papers here

reviewed by Evolve

first-author

corresponding

—

works · OpenAlex

—

h-index · OpenAlex

Bibliometric figures come from OpenAlex (2026-10-08) 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: no publication-count or productivity metrics are available in this snapshot beyond the two reviewed papers. With a fuller bibliographic record, this axis would measure total output, venue mix, and yearly consistency.

Impact—

Not yet scored: citation counts, h-index, and citation-normalized influence are missing from the current data. This axis would measure how widely the work on GUI agents is cited and taken up by other groups once citation data is attached.

Focus0/100

Radar reports a focus score of 0/100, but the underlying field distribution is an even 0.5/0.5 split between "artificial intelligence (GUI agents / vision-language models)" and "machine learning (vision-language GUI agents)". These two classifier labels are near-duplicates, both pointing at vision-language GUI agents, so the low score appears to reflect the even split across two fine-grained labels rather than real topical breadth. Across the two catalog papers, the substantive focus sits tightly on GUI agent training and navigation.

Collaboration63/100

Collaboration scores 63/100. Across the two reviewed papers, Huawen Shen worked with 23 distinct coauthors, and the frequent-coauthor pool includes Yan Zhang, Pei Fu, Daiqing Wu, Ruoceng Zhang, Shaojie Zhang, Jiahui Yang, Yu Zhou, Can Ma, Zhenbo Luo, and Jian Luan. Both papers carry middle-position, non-corresponding authorship, a pattern consistent with contributor roles inside a large industrial research team rather than sole or senior leadership of the work.

Activity—

Not yet scored: the snapshot provides only two 2026 papers, with no multi-year publication series to measure cadence. This axis would measure recent publication frequency and sustained research activity once a complete yearly record is available.

Reproducibility38/100

Reproducibility scores 38/100, computed as a mean of 38 across 2 reviewed papers. Evolve rates GUI-SD-v2's reproducibility readiness as medium and MaP's as low. Both verdicts converge on the same gap: the results are described as encouraging — MaP showing 2-3 point gains over mixture training on five benchmarks — but neither paper reports variance or significance tests, which limits confidence in the reported improvements.

Research summary

Huawen Shen is a researcher at Tencent (China) whose cataloged work centers on vision-language GUI agents — models that operate graphical user interfaces. Evolve's catalog holds two 2026 arXiv papers with Shen as a middle-position, non-corresponding author. The first, MaP, frames omni-dimensional GUI agent navigation as masked trajectory prediction, unifying navigation tasks within vision-language models and using LoRA adapters to ease gradient conflict; it reports 2-3 point gains over mixture training across five benchmarks. The second, GUI-SD-v2, extends on-policy self-distillation to multi-turn GUI agents through two-stage privileged follow-up optimization and selective step-wise distillation, with evaluation keyed to AndroidWorld and MobileWorld. Evolve's reviews rate reproducibility readiness as medium (GUI-SD-v2) and low (MaP), a mean reproducibility score of 38/100, and both verdicts note that the results look encouraging but lack variance or significance testing. Radar dimensions place collaboration at 63/100 across 23 distinct coauthors and reproducibility at 38/100, while output, impact, and activity axes remain unscored for lack of data. The profile is name-resolved from public bibliographic data.

Collaboration pattern

Both catalog papers carry large author lists — 23 distinct coauthors in total — with Huawen Shen in a middle, non-corresponding position, a pattern typical of team-based industrial research at Tencent. The recurring coauthor pool includes Yan Zhang, Pei Fu, Daiqing Wu, Ruoceng Zhang, Shaojie Zhang, Jiahui Yang, Yu Zhou, Can Ma, Zhenbo Luo, and Jian Luan. The radar collaboration score is 63/100. The source list contained a duplicate entry and Shen's own name, both removed here.

Research keywords

GUI agentson-policy self-distillationprivileged informationmulti-turn interactionAndroidWorldMobileWorldmulti-task learningmasked trajectory predictionvision-language modelsLoRA adaptersgradient conflict

Highlights

Extending on-policy self-distillation to multi-turn GUI agents
GUI-SD-v2 uses two-stage privileged follow-up optimization and selective step-wise distillation so a GUI agent can learn from its own interactions; Evolve's verdict calls the results encouraging while noting that no variance or significance tests are reported.
Masked trajectory prediction unifies GUI agent navigation (MaP)
MaP reframes omni-dimensional GUI navigation as masked trajectory prediction within vision-language models, using LoRA adapters to manage gradient conflict; it reports solid 2-3 point gains over mixture training across five benchmarks, though without significance tests.
Evaluation anchored in AndroidWorld and MobileWorld multi-turn settings
The GUI-SD-v2 paper's keyword set anchors it in multi-turn interaction, with AndroidWorld and MobileWorld named among its evaluation environments; Evolve assigns the paper a medium reproducibility readiness.

Timeline

2026The arXiv paper "Towards Omni-dimensional GUI Agent Navigation with Masked Trajectory Prediction" (MaP) enters the Evolve catalog, with Shen listed as a middle-position, non-corresponding author.
2026The arXiv paper "Learn How to Act from Your Own Interactions: On-Policy Self-Distillation for GUI Agents" (GUI-SD-v2) is cataloged, again with middle-position authorship.
2026Evolve reproducibility reviews cover both 2026 catalog papers, yielding a mean reproducibility score of 38/100 with readiness rated medium and low.

Papers in the Evolve catalog

Frequent co-authors

Yan Zhang×1 Pei Fu×1 Daiqing Wu×1 Huawen Shen×1 Ruoceng Zhang×1 Shaojie Zhang×1 Jiahui Yang×1 Yu Zhou×1 Can Ma×1 Zhenbo Luo×1 Jian Luan×1 Yan Zhang×1

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 OpenAlex-based public data only; no citation counts, h-index, or publication totals were available, so the output, impact, and activity axes are unscored.
The catalog covers only two 2026 arXiv preprints with middle-position authorship, so this profile likely under-represents the full publication record.
"Huawen Shen" is name-resolved from public bibliographic data; the name is not rare, so identity should be verified before reuse.
Field labels are classifier-generated and near-duplicate, which likely depresses the focus score to 0; the frequent-coauthor list contained a duplicate entry and the researcher's own name, both removed.