Baidu researcher bridging vision-language adaptation (PEFT layer selection) and tool-augmented long-video agents
2
papers here
2
reviewed by Evolve
0
first-author
0
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—
No publication-volume metrics are available in the current record, so this axis is unscored. Output would measure the quantity and consistency of papers over time — cumulative counts and annual productivity — data not present here.
Impact—
Impact cannot be scored because citation counts, h-index, and influence indicators are absent from this record. The axis would measure how widely the work is cited and used — citation totals, h-index, and downstream adoption — none of which are currently available.
Focus0/100
The focus score is 0, reflecting an even split of the reviewed output across two fields: computer vision (video understanding / multimodal agents) at 0.5 and machine learning (vision-language models) at 0.5. Rather than concentrating on a single area, the two 2026 papers attack complementary sides of multimodal research — one selects vision-encoder layers for PEFT adaptation, the other builds a tool-using agent for long video understanding.
Collaboration51/100
Collaboration scores 51, based on 12 distinct coauthors across the reviewed record. The frequent-coauthor list — including Sen Yang, Boqiang Duan, Jingdong Wang, and Hua Wu — points to work inside a larger Baidu research group. On both cataloged 2026 papers Liu is not the corresponding author (last author on one, middle author on the other), consistent with a contributing role in team-scale projects rather than solo-led studies.
Activity—
Activity data is insufficient to score this axis. It would capture recent research tempo — publications per year, preprint cadence, and continued engagement — but the record only shows two 2026 papers, not enough to characterize an ongoing pattern.
Reproducibility55/100
Reproducibility scores 55, the mean readiness across 2 assessed papers, each rated medium. Evolve's verdicts identify concrete gaps: the PEFT layer-selection paper never quantifies its core weight-statistics correlation, and the video-agent paper's ablations and baseline comparability need scrutiny — promising results, but replication would require careful re-verification.
Research summary
Jie Liu is a researcher at Baidu (China) whose reviewed work to date sits at the intersection of two areas in roughly equal measure: computer vision focused on video understanding and multimodal agents, and machine learning centered on vision-language models — the bibliometric field distribution assigns 0.5 to each. The Evolve catalog contains two 2026 arXiv papers. The first, Pre-PEFT Probing, where Liu is the last author, proposes using weight statistics and perturbation robustness to select which vision-encoder layers to adapt with parameter-efficient fine-tuning; Evolve notes the evidence is broad but the core correlation is never quantified. The second, an online video agent harness for long video understanding, where Liu is a middle author, reports a 60-plus-tool agent matching frontier large multimodal models at roughly 50k context, with strong efficiency results while ablation and baseline comparability need scrutiny. Both papers carry a medium reproducibility readiness, averaging 55 across 2 papers. Liu works within a sizeable team of 12 distinct coauthors and is not the corresponding author on either cataloged paper. As Jie Liu is a common name, this profile is name-resolved from public bibliographic data.
Collaboration pattern
Liu's reviewed record shows 12 distinct coauthors, with a frequent-coauthor group including Sen Yang, Boqiang Duan, Jing Yang, Weihao Bo, Boyuan Tong, Ze Feng, Wenkang Zhang, Jingdong Wang, Hua Wu, Qingtao Xia, Jiahua Bao, and Siyao Cheng — a pattern consistent with team-based industrial research at Baidu. Across the two 2026 papers Liu holds neither first nor corresponding authorship (last on the PEFT paper, middle on the video-agent paper), suggesting a contributing specialist role within larger multi-author efforts.
Research keywords
Highlights
Weight statistics and perturbation robustness for PEFT layer selection
Online 60+-tool video agent matching frontier LMMs at ~50k context
Timeline
- 2026Pre-PEFT Probing released on arXiv: weight statistics and perturbation robustness guide layer selection for PEFT in VLM vision encoders (last author).
- 2026Online Video Agent Harness for Long Video Understanding released on arXiv: a 60+-tool agent reported to match frontier LMMs at ~50k context (middle author).
- 2026Both cataloged papers fall in 2026, splitting evenly between VLM adaptation and video-agent research, each rated medium for reproducibility readiness.
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 signals are computed from OpenAlex-derived data only; no citation counts, h-index, or publication totals were available for this record.
- Jie Liu is a common name and no ORCID is attached, so this profile is name-resolved from public bibliographic data and identity disambiguation is uncertain.
- Only 2 reviewed papers exist in the Evolve catalog for this researcher, both 2026 arXiv preprints; peer-reviewed venue output may be missing.
- Output, impact, and activity radar axes are unscored because the underlying bibliometric data is insufficient.