UVA researcher in efficient LLM inference: evidence-preserving compression and training-free acceleration
2
papers here
2
reviewed by Evolve
0
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—
Not yet scored: only two reviewed 2026 arXiv papers are in scope, and aggregate output metrics (publication count over time, venues) are unavailable. This axis would measure publication volume, venue quality, and output consistency across the career.
Impact—
Not yet scored: no citation counts or downstream-uptake records are linked to this profile. This axis would measure citations, adoption of MORSE and WISE, and influence on efficient-inference research.
Focus0/100
Evolve records a focus score of 0, derived from an even field distribution: 50% natural language processing and 50% machine learning (efficient inference for language models) across the two reviewed 2026 papers. Thematically the work is more unified than the score suggests — both papers target inference-time efficiency for LLMs, one via evidence-preserving compression and one via training-free recurrent-model acceleration — but a two-paper sample makes concentration unreliable.
Collaboration32/100
The collaboration score is 32, computed from 4 distinct coauthors across the reviewed papers. Recurring collaborators are Ke Wan, Yifan Wang, and Liheng Lai. On both 2026 papers Chen Chen sits in the last author position without being flagged as corresponding author — a pattern consistent with a senior, supervising role within a small, stable core team rather than a broad external network.
Activity—
Not yet scored: the snapshot covers only two 2026 papers, with no publication-date series to measure cadence. This axis would measure yearly output rhythm and research continuity over time.
Reproducibility100/100
Reproducibility is the strongest recorded axis: a perfect score of 100, computed as a mean of 100 across 2 reviewed papers (reproPapers = 2), both rated high readiness. Evolve's verdicts note concrete, bounded evidence — MORSE reports evidence-retention gains in all 24 settings with mixed QA gains stated openly, and WISE is training-free with its limitation to two small backbones disclosed — a pattern of transparent claims that supports re-runnability.
Research summary
Chen Chen is a researcher at the University of Virginia whose reviewed work centers on making large language models cheaper and more faithful at inference time. Two 2026 arXiv papers, both with Chen Chen in the last author position, define this early-stage profile. The first, MORSE, studies multi-context compression and finds that the order in which contexts are presented changes what an LLM compressor retains; its reverse-scored anchor combined with search raises evidence retention in all 24 tested settings, though downstream QA gains are mixed. The second, WISE, examines recurrent-depth language models and shows that attention routing stabilizes about five steps before the underlying representations do; exploiting this early-stabilizing support lets WISE reuse a compact working set training-free, preserving QA performance through 2K steps, with results demonstrated on two small backbones only. Bibliometrically, Evolve records a perfect reproducibility score (mean 100 across 2 papers, both rated high readiness), a focus score of 0 driven by an even 50/50 field split between natural language processing and machine learning (efficient inference for language models), and a collaboration score of 32 from 4 distinct coauthors, including recurring collaborators Ke Wan, Yifan Wang, and Liheng Lai. Output, impact, and activity metrics are not yet available, and 'Chen Chen' is a common name, so this profile is name-resolved from public bibliographic data.
Collaboration pattern
The reviewed work comes from a compact core team: 4 distinct coauthors in total, with Ke Wan, Yifan Wang, and Liheng Lai recurring across outputs. Chen Chen holds the last author position on both 2026 papers without a corresponding-author flag, consistent with a senior supervising role. The collaboration score of 32 reflects this small, stable group; the two papers split evenly between NLP compression and efficient recurrent-LM inference, and no external or cross-institution partners appear in the available data.
Research keywords
Highlights
MORSE: reverse-scored context ordering preserves evidence under compression
WISE: early-stabilizing attention routing enables training-free acceleration
Timeline
- 2026Published MORSE on arXiv (last-author position): reverse-scored multi-context ordering for evidence-preserving LLM compression, with retention gains in all 24 settings.
- 2026Published WISE on arXiv (last-author position): showed recurrent-LM attention routing stabilizes within about 5 steps, and reused that support for training-free working-set inference.
- 2026Both papers entered the Evolve review catalog with high reproducibility readiness, yielding a perfect reproducibility mean of 100 across 2 papers.
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.
- Bibliometric scores are computed from a narrow snapshot: only 2 reviewed 2026 arXiv papers; output, impact, and activity metrics are missing.
- Reproducibility score of 100 rests on only 2 papers, so the perfect rating may not generalize.
- 'Chen Chen' is a common name; this profile is name-resolved from public bibliographic data tied to University of Virginia, and no ORCID is available for verification.
- Both papers are arXiv preprints; peer-review status and venue acceptance are unknown.