Jian Liu

Tongji University

Unclaimed · AI-structured profileOpenAlex

Tongji University researcher working across speaker verification and vision-language-action robot learning.

papers here

reviewed by Evolve

first-author

corresponding

—

works · OpenAlex

—

h-index · OpenAlex

Bibliometric figures come from OpenAlex (2026-10-14) 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 data is insufficient for scoring: this profile currently tracks only the 2 arXiv preprints in the Evolve catalog, with no full publication count, venue history, or yearly totals. This axis would measure total and per-year publication volume once a complete bibliographic record is available.

Impact—

Impact data is not yet available: no citation counts, h-index, or per-paper citation metrics were provided for these two 2026 preprints. This axis would measure citations and community influence of the work once such metrics exist.

Focus0/100

The focus score is 0, grounded in the raw field distribution: the cataloged papers split evenly, 50% in speech processing / speaker recognition and 50% in robot learning / vision-language-action models. The two 2026 preprints sit in distinct communities — MECT on VoxCeleb speaker verification and GeoAAC on flow-matching VLA policies — so no dominant single research direction can yet be identified from the available data.

Collaboration51/100

The collaboration score is 51, based on 12 distinct coauthors in the bibliometric snapshot. Both cataloged papers list Jian Liu in a middle author position and neither as corresponding author, consistent with contributing inside larger project teams rather than leading them. Recurring collaborators include Yu Zheng, Jinghan Peng, ChangHao Zhang, Weiqiang Wang, Xin Chen, Sen Chen, Yujuan Ding, Guoqing Wang, Wei Ye, Heng Tao Shen and Yi Bin.

Activity—

Activity data is insufficient for scoring: with only two 2026 preprints and no multi-year publication record, the recent output cadence cannot be assessed. This axis would measure publication frequency and sustained research engagement over time once a longer record is available.

Reproducibility55/100

Reproducibility averages 55 across 2 scored papers, and both are rated medium readiness. Evolve's reviews identify concrete gaps: MECT's state-of-the-art VoxCeleb minDCF results come from a single run and are untested for statistical significance, while GeoAAC reports consistent but partly small gains with no significance tests and no code release. The moderate score reflects complete method descriptions held back by missing artifacts rather than unclear methods.

Research summary

Jian Liu is a researcher affiliated with Tongji University whose visible recent output spans two distinct areas: speaker verification in speech processing and vision-language-action (VLA) models in robot learning. The Evolve-reviewed catalog contains two 2026 arXiv preprints, both listing Jian Liu as a middle author and neither as corresponding author, indicating contribution within larger project teams rather than sole leadership. The first, MECT, presents a mixture-of-experts CNN-Transformer network for speaker verification, claiming state-of-the-art VoxCeleb minDCF results and 100 ms streaming inference; Evolve rates its reproducibility readiness as medium, noting single-run results without significance testing. The second, GeoAAC, proposes a training-free adaptive action chunking method derived from denoising-trajectory geometry in flow-matching VLA policies, with consistent but partly small gains and no code release. Bibliometric signals reflect this split profile: field distribution is an even 50/50 between the two areas, the focus score is 0, collaboration scores 51 with 12 distinct coauthors, and reproducibility averages 55 across two scored papers. Output, impact, and activity metrics are not yet available and remain unscored.

Collaboration pattern

Jian Liu works within multi-author teams: both 2026 cataloged papers are middle-author, non-corresponding contributions. The snapshot records 12 distinct coauthors, with recurring collaborators including Yu Zheng, Jinghan Peng, ChangHao Zhang, Weiqiang Wang, Xin Chen, Sen Chen, Yujuan Ding, Guoqing Wang, Wei Ye, Heng Tao Shen and Yi Bin. The same collaborator pool spans both research lines — speech and robot learning — a pattern consistent with shared team structures, though the exact grouping cannot be confirmed from the available data.

Research keywords

speaker verificationmixture of expertsCNN-Transformer architecturestreaming inferenceVoxCeleb benchmarksvision-language-action policiesaction chunkingadaptive action horizonflow matchingdenoising trajectory geometrypredictive uncertaintycausal retraining

Highlights

MECT: mixture-of-experts CNN-Transformer for speaker verification
Claims the first fully supervised MoE-based speaker verification model, reporting state-of-the-art VoxCeleb minDCF and 100 ms streaming inference supported by causal retraining. Evolve judges the results strong but from a single run and untested for statistical significance, with medium reproducibility readiness.
GeoAAC: training-free adaptive action chunking for VLA policies
Proposes adaptive action-horizon selection derived from the geometry of denoising trajectories in flow-matching VLA policies, requiring no retraining and leveraging predictive uncertainty. Evolve finds the gains consistent but partly small, with no significance tests or code release and medium reproducibility readiness.

Timeline

2026Co-authored the arXiv preprint MECT, a mixture-of-experts CNN-Transformer network for speaker verification with claimed state-of-the-art VoxCeleb results, in a middle author position.
2026Co-authored the arXiv preprint GeoAAC, proposing geometry-based adaptive action chunking from denoising trajectories for VLA policies, also in a middle author position.
2026Both preprints entered the Evolve catalog in the same year, splitting evenly across speech processing / speaker recognition and robot learning / vision-language-action models.

Papers in the Evolve catalog

Frequent co-authors

Yu Zheng×1 Jinghan Peng×1 ChangHao Zhang×1 Jian Liu×1 Weiqiang Wang×1 Xin Chen×1 Sen Chen×1 Yujuan Ding×1 Guoqing Wang×1 Wei Ye×1 Heng Tao Shen×1 Yi Bin×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 come from a public bibliometric snapshot; output, impact, and activity metrics are missing, so those radar axes are unscored.
'Jian Liu' is a common name; this profile is name-resolved from public bibliographic data and identity disambiguation is not guaranteed.
Both cataloged papers are 2026 arXiv preprints; their peer-reviewed publication status is unknown.
Jian Liu appears as a middle author on both papers, so per-paper contribution and leadership roles cannot be assessed from the available data.