University of Virginia researcher in efficient LLM inference — compression, context ordering, recurrent language models
2
papers here
2
reviewed by Evolve
2
first-author
0
corresponding
—
works · OpenAlex
—
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-02) 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—
This axis would measure publication volume and authorship roles over time. The current snapshot lacks those counts — only the two cataloged 2026 papers are on record — so no output score is computed yet.
Impact—
Impact would track citations, downstream adoption, and influence of the reviewed work. No citation counts or influence metrics are available in the provided data, so this axis is left unscored rather than estimated.
Focus0/100
Focus scores 0 on the 0–100 scale, and the underlying field distribution is an exactly even two-way split — 0.5 natural language processing, 0.5 machine learning (efficient inference for language models). In practice the two halves address one theme, efficient inference for large language models: compression and context ordering in MORSE, recurrent-architecture acceleration in WISE. The score should be read together with this narrow two-field distribution.
Collaboration32/100
The collaboration axis scores 32 on the 0–100 scale, based on 4 distinct coauthors in the record. The recurring names are Chen Chen, Yifan Wang, and Liheng Lai (Ke Wan's own name also appears in the frequent-coauthor list). Both cataloged papers are first-authored by Ke Wan without a corresponding-author flag, consistent with a hands-on lead role inside a small, stable team rather than a large distributed collaboration.
Activity—
Activity would measure publication cadence — papers per year, recency, and sustained output across years. The provided snapshot covers only two 2026 papers and offers no multi-year publication history, so this axis cannot be scored yet.
Reproducibility100/100
Reproducibility scores 100, the top of the 0–100 scale, with a mean readiness of 100 computed across the 2 cataloged papers; both MORSE and WISE carry a high readiness rating from Evolve's review. The reviewed evidence is concrete — MORSE is evaluated across 24 settings and WISE on two small recurrent backbones — which supports independent reimplementation, though the tested scope remains limited.
Research summary
Ke Wan is a researcher at the University of Virginia whose published work to date sits squarely on efficient inference for large language models. The Evolve catalog contains two 2026 arXiv papers, both with Ke Wan as first author. The first, MORSE, studies multi-context compression and shows that the order in which contexts are presented changes what an LLM's compression keeps; its reverse-scored anchor plus search procedure raises evidence retention across all 24 tested settings, though question-answering gains remain mixed. The second, WISE, examines recurrent-depth language models and finds that attention routing stabilizes roughly five steps before representations do; WISE exploits this with training-free working-set reuse, preserving QA performance through 2K, though evaluation covers only two small backbones. Bibliometrically, the record splits evenly between natural language processing and machine learning for efficient inference (0.5 each). Both cataloged papers carry high reproducibility readiness, with a mean of 100 across 2 papers. The visible collaboration footprint is compact — 4 distinct coauthors, with Chen Chen, Yifan Wang, and Liheng Lai recurring. Output, impact, and activity axes lack sufficient bibliometric data and are not scored.
Collaboration pattern
Ke Wan's visible collaboration footprint is compact: 4 distinct coauthors appear in the record, and the frequent-coauthor list centers on Chen Chen, Yifan Wang, and Liheng Lai (Ke Wan's own name also appears in that list). Both cataloged papers are first-authored without a corresponding-author flag, consistent with a hands-on lead role inside a small, stable team. No large multi-institution network is visible in the current snapshot.
Research keywords
Highlights
MORSE: evidence retention holds across all 24 settings
WISE: routing stabilizes ~5 steps early, enabling training-free reuse
Timeline
- 2026First-author arXiv paper MORSE on multi-context ordering via reverse scoring for evidence-preserving compression.
- 2026First-author arXiv paper WISE shows recurrent-LM attention routing stabilizes about 5 steps early, enabling training-free working-set inference.
- 2026Both papers enter the Evolve catalog with high reproducibility readiness (mean 100 across 2 papers), and the profile's focus splits evenly between NLP and efficient inference.
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.
- Profile is name-resolved from public bibliographic data; Ke Wan is a common name, so identity rests only on the cataloged papers and the listed affiliation.
- Dimensions are computed only from the provided bibliometric snapshot; no citation counts, h-index, or venue list was supplied, so output, impact, and activity remain unscored.
- The reviewed catalog holds only two papers, both 2026 arXiv preprints; peer-reviewed venue acceptance is not on record.
- Affiliation country and ORCID are missing, and the WISE results were reported on two small backbones only, limiting generalization.