Evolve.Pub
Sign inScan your paper
WW

Wei Wang

Xi'an Jiaotong University · China

Unclaimed · AI-structured profileOpenAlex

Xi'an Jiaotong University researcher in LLM agent security, adversarial robustness and safety benchmark evaluation

5

papers here

5

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: insufficient dataOutputImpact: insufficient dataImpactFocus: 0/100FocusCollaboration: 78/100CollaborationActivity: 64/100ActivityReproducibility: 48/100Reproducibility

Output—

Output is not yet scorable: the snapshot lacks a reliable total-work count for this author, recording only 8 works in the recent three-year window and 8 works in 2026. A complete record would measure cumulative publication volume and yearly productivity.

Impact—

Impact cannot be scored: h-index and i10 fields read zero and no citation aggregates are populated in this record. The axis would measure citation accumulation, h-index trajectory, and uptake of the group's benchmarks and attacks by follow-up work.

Focus0/100

The focus score sits at 0/100, matching an unusually even field distribution: NLP, AI safety/agent robustness, AI security, graph representation learning, and agent-security software engineering each hold 0.2 weight. Topic data shows one mild center of gravity — Adversarial Robustness in Machine Learning accounts for 4 topic mentions versus 1 each for complex networks, sensor localization, software engineering, and 3D shape modeling — so the portfolio is broad rather than undisciplined, but currently lacks a dominant single track.

Collaboration78/100

Collaboration scores 78/100 on the strength of 50 distinct coauthors in the record, with 12 recurring collaborators including Yifan Wang, Zimu Wang, Suliu Qin, Changyu Zeng, Tong Chen, Siqi Chen, Yijie Lin, Lingyu Jiang, Jionglong Su, Yushan Pan, Haiyang Zhang, and Qiaoyu Tan. Five papers spanning agent security, robustness evaluation, NLP, and graph learning within a single year imply several parallel working teams rather than one fixed group.

Activity64/100

Activity scores 64/100. The record shows 8 works in the recent three-year window and 8 works in 2026 alone, and all five reviewed papers carry 2026 arXiv dates. The signal is a concentrated recent burst: every catalogued output lands in a single year, which supports ongoing momentum but leaves the longer-term cadence unmeasured in this snapshot.

Reproducibility48/100

Reproducibility scores 48/100, the mean readiness across the 5 reviewed papers. Four papers — AgentLeak, SWE-Test, the deep-search poisoning benchmark, and SinoGlyphBench — rate medium readiness, with reviewers noting gaps such as single-run trials, LLM-judge scoring, and small recovery subsets. The granular-ball hypergraph paper rates low, with weak statistics, spec gaps, and template leftovers cited against rigor.

Research summary

Wei Wang, profiled via OpenAlex bibliographic records (a common name, resolved bibliographically rather than by ORCID), is a researcher at Xi'an Jiaotong University whose reviewed work centers on the security and robustness of LLM agents. Five 2026 arXiv papers sit in the Evolve catalog: AgentLeak, showing weaker agents can clone stronger LLM agents' capabilities with over 80% gap recovery (Wang as last author); SWE-Test, recasting vulnerability discovery as branch-targeted input prediction; a hierarchical web-evidence-poisoning benchmark for deep-search agents; SinoGlyphBench, a matched-pair diagnostic showing glyph obfuscation degrades Chinese moderation across 12 models; and a granular-ball hypergraph learning method claiming state of the art on 7 benchmarks (also last author). Topic records weight Adversarial Robustness in Machine Learning highest (4 of 8 topic mentions), while field distribution splits evenly at 0.2 across NLP, agent security, agent robustness, graph learning, and agent-security software engineering. The record lists 50 distinct coauthors, and reproducibility readiness across the five reviewed papers averages 48/100, with four rated medium and one low.

Collaboration pattern

Records list 50 distinct coauthors with 12 recurring collaborators, among them Yifan Wang, Zimu Wang, Suliu Qin, Changyu Zeng, Tong Chen, Siqi Chen, Yijie Lin, Lingyu Jiang, Jionglong Su, Yushan Pan, Haiyang Zhang, and Qiaoyu Tan. Five 2026 papers spanning agent security, robustness evaluation, Chinese NLP, and graph learning suggest small parallel teams around a shared benchmark-and-attack methodology rather than one fixed lab group; authorship positions range from middle to last.

Research keywords

LLM agent securityadversarial robustnesscapability extraction attacksweb evidence poisoningred-team benchmarkingcontent moderationChinese NLPLLM safety evaluationvulnerability discoverygraph representation learninghypergraph neural networksgranular-ball computing

Highlights

Timeline

  1. 2026AgentLeak appears on arXiv: black-box capability cloning of LLM agents with over 80% gap recovery, Wang as last author.
  2. 2026SWE-Test benchmark posted, recasting LLM-driven vulnerability discovery as branch-targeted input prediction.
  3. 2026Robustness evaluation of deep-search agents under hierarchical web evidence poisoning enters the catalog.
  4. 2026SinoGlyphBench diagnostic shows glyph-level obfuscation degrading Chinese moderation across 12 models.
  5. 2026Granular-ball multi-granularity hypergraph learning method posted, claiming SOTA on 7 benchmarks with Wang as last author.

Papers in the Evolve catalog

Frequent co-authors

Yifan Wang×1Zimu Wang×1Suliu Qin×1Changyu Zeng×1Tong Chen×1Siqi Chen×1Yijie Lin×1Lingyu Jiang×1Jionglong Su×1Yushan Pan×1Haiyang Zhang×1Qiaoyu Tan×1

Is this you? Claim this profile

Enter the email you published with. We'll send a one-time claim link; after claiming you can edit the introduction or hide this page. Requests from a mismatched address go to manual review.

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.

  • Identity is name-resolved from public OpenAlex bibliographic data; 'Wei Wang' is a common name and no ORCID is linked, so attribution carries inherent ambiguity.
  • Bibliometrics come from OpenAlex only; h-index and i10 fields read zero and citation aggregates are unpopulated, so the output and impact axes are unscored.
  • All five reviewed papers are 2026 arXiv preprints; peer-reviewed venue status and acceptance outcomes are not reflected in this snapshot.
  • Publication history in the record covers only 2026 (8 works), so long-term productivity and career trajectory cannot be assessed.
Wei Wang — Researcher Profile | Evolve.Pub