Tongji University first-author working on LLM safety interpretability and vision-language-action robot learning.
2
papers here
2
reviewed by Evolve
2
first-author
0
corresponding
—
works · OpenAlex
0
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—
Data insufficient: this axis measures publication volume and productivity over time, but the snapshot holds only two 2026 preprints (recent3yWorks = 2) with no full output history, so no score is computed.
Impact—
Data insufficient: this axis would capture citation-based influence (citations, h-index, i10-index); the snapshot lists both indices as zero with no citation counts, so impact is not yet measurable and stays unscored.
Focus0/100
Scores 0 for concentration, reflecting a perfectly even split in the field distribution: 0.5 machine learning (LLM safety / interpretability) and 0.5 robot learning / vision-language-action models. The five indexed topics — topic modeling, reinforcement learning in robotics, robot manipulation and learning, AI-based problem solving and planning, and anomaly detection — each carry a count of 1, confirming work spread across two areas rather than one sustained specialization.
Collaboration48/100
Scores 48, based on 10 distinct coauthors in the recent window. The frequent-coauthor list spans both research lines and includes Gil Kur, Alexander Shevchenko, Andreas Krause, Sen Chen, Yujuan Ding, Jian Liu, Guoqing Wang, Wei Ye, Heng Tao Shen, and Yi Bin. Both catalog papers are first-author and non-corresponding, so the record shows a leading contributor role within teams, while the moderate score reflects a limited cumulative collaboration footprint.
Activity32/100
Scores 32. The publication window is narrow: countsByYear records 2 works in 2026 and none in earlier years, and recent3yWorks equals 2. Both items are first-author arXiv preprints from 2026, so current activity is concentrated in a single recent year; the low score reflects the short indexed history rather than a measured slowdown.
Reproducibility78/100
Scores 78, the mean reproducibility-readiness across the 2 reviewed papers (reproMean 78, reproPapers 2). "Local Sparsity Enables Unsupervised LLM Safety Detection" is rated high, backed by theory and six-model validation with strong ablations; "GeoAAC" is rated medium, with consistent but partly small gains and no significance tests or code release. This is the strongest dimension in the profile.
Research summary
Xin Chen is a researcher at Tongji University whose published work to date spans two distinct lines of machine learning research, each represented by a first-author 2026 preprint in the Evolve catalog. The first line, in LLM safety and interpretability, contributes "Local Sparsity Enables Unsupervised LLM Safety Detection", which proposes locally-masked sparse-autoencoder anomaly detection for identifying unsafe behavior in large language models, supported by theoretical grounding and validation across six models; Evolve rates its reproducibility readiness as high. The second line, in robot learning, contributes "GeoAAC", a training-free method that adapts action chunking horizons in vision-language-action policies using the geometry of denoising trajectories from flow-matching policies; Evolve notes consistent gains but also the absence of significance testing and code release. Per the bibliometric snapshot, the two papers split evenly (0.5/0.5) between the two fields, ten distinct coauthors appear in the recent window, and the mean reproducibility score across the two reviewed papers is 78. OpenAlex-derived aggregates show two works in 2026 and h-index and i10-index values of zero, indicating an early-stage citation record rather than measured impact. The profile is name-resolved from public bibliographic data, as "Xin Chen" is a common name.
Collaboration pattern
The recent record lists 10 distinct coauthors, and the frequent-coauthor set spans both research lines: Gil Kur, Alexander Shevchenko, Andreas Krause, Sen Chen, Yujuan Ding, Jian Liu, Guoqing Wang, Wei Ye, Heng Tao Shen, and Yi Bin. Both catalog papers are first-author and non-corresponding, suggesting a leading-contributor role within teams rather than a corresponding-author position; the collaboration score is 48.
Research keywords
Highlights
Unsupervised LLM safety detection via local sparsity
Training-free adaptive action chunking for VLA policies
Timeline
- 2026First-author preprint "Local Sparsity Enables Unsupervised LLM Safety Detection" appears, opening the LLM safety and interpretability line.
- 2026First-author preprint "GeoAAC" appears, establishing the robot-learning line on adaptive action chunking in vision-language-action policies.
- 2026Both papers enter the Evolve catalog with reproducibility ratings of high and medium, giving a mean reproducibility score of 78 across 2 papers; OpenAlex logs 2 works for 2026 with h-index and i10-index at 0.
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.
- "Xin Chen" is a common name; this profile is name-resolved from public bibliographic data and no ORCID is linked.
- All bibliometric aggregates come from OpenAlex only; no other citation source is included.
- Citation-impact data is insufficient: h-index and i10-index are zero in the snapshot, reflecting missing indexed citations rather than measured impact.
- Only two papers are reviewed, both 2026 arXiv preprints; peer-reviewed venues, career history, and awards are outside the data scope.