LLM researcher at Southeast University spanning safety red-teaming, efficient inference, and multilingual evaluation
3
papers here
1
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
0
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 metrics are not yet available: the OpenAlex record behind this profile shows no usable publication totals or citation-based output measures, so this axis cannot be scored. It would measure paper volume, venue spread, and yearly productivity.
Impact—
Impact cannot be assessed yet: citation counts, h-index, and i10-index are missing from the underlying OpenAlex profile, likely because author disambiguation is incomplete. This axis would measure citation influence and scholarly uptake of the 2026 preprints.
Focus100/100
Focus scores 100/100: the OpenAlex topic profile is tightly concentrated — Topic Modeling (2 works) and Natural Language Processing Techniques (2 works) dominate, with single works in Advanced Neural Network Applications, Ethics and Social Impacts of AI, and Adversarial Robustness. All three reviewed 2026 papers sit on one theme, evaluating and constraining large language models, from jailbreak red-teaming to inference efficiency to pragmatics benchmarking.
Collaboration58/100
Collaboration scores 58/100. The OpenAlex record lists 17 distinct coauthors in the recent window, and 12 recurring collaborators — including Yao Huang, Yinpeng Dong, Junyi Jessy Li, and Geyang Guo — appear in the Evolve network. On the three reviewed papers Xu sits last author twice and middle once, none as corresponding author, a pattern consistent with a senior or supervisory contributor within small-to-mid-size teams.
Activity40/100
Activity scores 40/100. countsByYear shows a single indexed year — 2026 with 3 works — and recent3yWorks is likewise 3, all arXiv preprints. Output is recent but sparse, so the score reflects a thin, front-loaded record rather than sustained multi-year productivity; earlier works may exist but are not captured in this OpenAlex profile.
Reproducibility55/100
Reproducibility scores 55/100, computed from one scored paper (reproMean 55, reproPapers 1); all three reviewed preprints carry a medium repro-readiness rating. Evolve notes no code release for HeadWiseKV, plus adapted baselines and a single-GPU setup, while BLUEPRINT and VakyArth are well-validated but bounded — query-cost framing and small per-cell samples temper replication confidence.
Research summary
Wei Xu (Southeast University) is an LLM researcher whose indexed profile is early-stage but whose 2026 output spans three adjacent fronts of large language model research. On safety, the BLUEPRINT paper introduces worldview simulation to amplify psychologically grounded persuasion in multi-turn jailbreaking, reaching near-ceiling attack success at an average of 2.46 queries, with Xu in the last author position. On systems, HeadWiseKV proposes budgeted per-head KV cache residency for hybrid long-context models, retaining near-full quality at 69% cache retention, again with Xu as last author. On evaluation, VakyArth delivers the first Indic pragmatics benchmark — 578 natively authored items across four languages — where Xu is a middle author. OpenAlex topics mirror this spread: Topic Modeling (2) and Natural Language Processing Techniques (2) lead, with single entries in Ethics and Social Impacts of AI and Adversarial Robustness. The record shows 3 works in the last three years, 17 distinct coauthors, and a reproducibility mean of 55 over one scored paper. Citation-based output and impact metrics are not yet available, so positioning rests on the reviewed preprints.
Collaboration pattern
Wei Xu's OpenAlex record shows 17 distinct coauthors in the recent window. Twelve recur in the Evolve network, including Siyu Chen, Haoran Wang, Yao Huang, Yinpeng Dong, Xiaojian Li, Geyang Guo, Junyi Jessy Li, and Renjie Xie — spanning safety/red-teaming, systems, and multilingual NLP clusters. On the reviewed papers Xu takes the last author position twice and a middle position once, none as corresponding author, suggesting a senior-contributor pattern within small-to-mid-size teams rather than solo work.
Research keywords
Highlights
BLUEPRINT: worldview simulation drives multi-turn jailbreak persuasion
HeadWiseKV: per-head KV cache budgeting for hybrid long-context models
VakyArth: the first Indic pragmatics benchmark for LLMs
Timeline
- 2026Publishes BLUEPRINT, a worldview-simulation approach to psychologically grounded multi-turn jailbreak persuasion (last author; arXiv).
- 2026Publishes HeadWiseKV on budgeted per-head KV cache residency for hybrid long-context LLMs (last author; arXiv).
- 2026Contributes to VakyArth, the first Indic pragmatics benchmark for LLM evaluation (middle author; 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.
- "Wei Xu" is a common name; this profile is name-resolved from public bibliographic data (OpenAlex), and the listed country (US) was not reconciled with the Southeast University affiliation.
- Bibliometrics come from OpenAlex only; h-index and i10-index read 0 and the output and impact axes are marked insufficient, likely reflecting an incomplete or sparsely disambiguated author record.
- All reviewed papers are 2026 arXiv preprints; no peer-reviewed venue acceptance or citation uptake is observable yet.
- The reproducibility score rests on a single scored paper (mean 55); the other two carry only medium repro-readiness ratings.