Jun LI

China University of Geosciences

Unclaimed · AI-structured profileOpenAlex

Jun Li, China University of Geosciences — statistical learning for unmixing and test-time image restoration

papers here

reviewed by Evolve

first-author

corresponding

—

works · OpenAlex

—

h-index · OpenAlex

Bibliometric figures come from OpenAlex (2026-10-15) 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 cannot be scored yet: the provided data contains no publication counts, venue list, or productivity metrics. This axis would measure total papers, spread across venues, and yearly production volume once a fuller bibliographic record becomes available.

Impact—

Impact is not scored: the provided data includes no citation counts, h-index, or downstream-usage signals. This axis would measure how often and how widely the community cites and builds on this researcher's published work.

Focus0/100

The focus axis scores 0, and the underlying field distribution explains why: across the reviewed record, output splits exactly 50/50 between mathematical statistics / statistical machine learning for hyperspectral unmixing and computer vision for image restoration. With only two reviewed papers, one per field, there is not yet enough mass to call either a core; the split reads as genuine breadth or an early search for direction.

Collaboration50/100

Collaboration scores 50 on a 0-100 scale, with 11 distinct coauthors recorded across the reviewed record. The frequent-coauthor list is anchored by Yaowei Wang, who appears twice, alongside Tao Dai, Jinpeng Wang, Yanlong Guo, Zhaozhao Zeng, Kaihang Zheng, Hang Guo, Hongyu Chi, and Zimo Liu. Li held first authorship on one 2026 paper and a middle position on the other, consistent with both leading and contributing roles within the same small network.

Activity—

Activity is not scored: beyond two 2026 arXiv preprints, no publication cadence or recency series is available. This axis would measure annual output rhythm and whether recent years show sustained or accelerating research engagement.

Reproducibility78/100

Reproducibility scores 78, the mean across the two reviewed papers (reproPapers = 2). The first-author simplex-estimation paper rates high readiness: its non-asymptotic guarantees are explicit, though Evolve notes the certified regime dwarfs the experiments and Adam training is excluded. TTTIR rates medium — reported state-of-the-art gains, including 30.78 dB on LOL-v2-Real, still need independent verification. The score reflects strong methodological disclosure rather than fully confirmed results.

Research summary

Jun Li is a researcher at China University of Geosciences whose reviewed work spans two areas in equal measure: mathematical statistics and statistical machine learning applied to hyperspectral unmixing, and computer vision for image restoration — the profile's field distribution splits 50/50 between them. Two 2026 arXiv preprints are in the Evolve catalog. The first, with Li as first author, proposes a scalable minimum-volume simplex estimator built on neural implicit representations, claiming O(K^2) memory scaling to N=10^8 together with non-asymptotic sample-complexity analysis; Evolve judges the work rigorous but notes the certified regime dwarfs the experiments and that Adam-based training is excluded. The second, TTTIR, places Li in a middle authorship position and applies test-time training to instance-specific state evolution for low-light enhancement, deraining, and dehazing, reporting strong state-of-the-art gains — with some results flagged for verification. The record shows 11 distinct coauthors and a reproducibility mean of 78 across the two papers, one rated high readiness and one medium. Output, impact, and activity metrics are not yet computable, so this should be read as an early-stage, name-resolved profile built from a small, very recent footprint rather than a settled publication record.

Collaboration pattern

With 11 distinct coauthors across two 2026 papers, Jun Li works in a moderately sized team (collaboration score 50). The frequency list is headed by Yaowei Wang, who appears twice, followed by Tao Dai, Jinpeng Wang, Yanlong Guo, Zhaozhao Zeng, Kaihang Zheng, Hang Guo, Hongyu Chi, and Zimo Liu. Authorship patterns span both modes: first author on the simplex-estimation paper and a middle position on TTTIR, suggesting comfort with both leading and supporting roles.

Research keywords

simplex estimationminimum-volume estimationhyperspectral unmixingneural implicit representationssample complexitynon-asymptotic analysistest-time trainingimage restorationlow-light enhancementimage derainingimage dehazingstate evolution

Highlights

First-author scalable simplex estimation with non-asymptotic theory
The 2026 first-author paper proposes a neural implicit simplex estimator for minimum-volume estimation in hyperspectral unmixing, claiming O(K^2) memory scaling to N=10^8 with non-asymptotic analysis. Evolve rates the work rigorous but flags that the certified regime dwarfs the experiments and Adam training is excluded; reproducibility readiness is high.
Test-time training for instance-specific image restoration
TTTIR, with Li in a middle authorship position, adapts restoration models at inference via test-time training with spatial-frequency target states, reporting strong state-of-the-art gains across low-light enhancement, deraining, and dehazing. Evolve flags several results — including 30.78 dB on LOL-v2-Real — as needing verification; readiness is medium.
A caution-forward early record
With a reproducibility mean of 78 over just two reviewed papers, this is a promising but unproven record: the theory-heavy paper excludes Adam training from its guarantees, and TTTIR's headline numbers await independent confirmation.

Timeline

2026Releases an arXiv first-author preprint on scalable minimum-volume simplex estimation with non-asymptotic analysis for hyperspectral unmixing.
2026Contributes as a middle author to TTTIR, an arXiv preprint applying test-time training to adaptive image restoration (low-light, deraining, dehazing).
2026Both papers enter the Evolve catalog in the same year, splitting evenly between statistical learning and computer vision, with 11 distinct coauthors on record.

Papers in the Evolve catalog

2026-09-22first author
Scalable Minimum-Volume Simplex Estimation with Non-asymptotic Analysis
2026-09-15
TTTIR: Unlocking Instance-Specific State Evolution via Test-Time Training for Image Restoration

Frequent co-authors

Yanlong Guo×1 Zhaozhao Zeng×1 Kaihang Zheng×1 Jun Li×1 Hang Guo×1 Hongyu Chi×1 Zimo Liu×1 Tao Dai×1 Jinpeng Wang×1 Yaowei Wang×1 Yaowei Wang×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.

Bibliometric scores derive from OpenAlex-based data only; no citation counts, h-index, or full publication list were available.
'Jun Li' is a very common name and no ORCID is provided; this profile is name-resolved from public bibliographic records, so homonym mixing is possible.
The evidence base is only two Evolve-reviewed arXiv preprints, both dated 2026; output, impact, and activity axes are marked as missing.
The reproducibility mean of 78 covers just the two reviewed papers; several TTTIR results, including 30.78 dB on LOL-v2-Real, are flagged for verification.