Researcher at Beijing Jiaotong University working across LLM value alignment and generalizable deepfake detection
2
papers here
2
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
—
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-16) 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—
Publication-output data (total counts, venue tiers, yearly volume) is not available in this snapshot, so the axis is left unscored. It would measure cumulative and per-year output across indexed sources, beyond the two 2026 arXiv preprints currently on record.
Impact—
Citation-based impact metrics (citation counts, h-index, citing fields) are absent from the current data, so this axis is not scored. It would measure how often the reviewed works are cited and by which research communities.
Focus0/100
The focus axis scores 0, consistent with the raw field distribution: an exact 50/50 split between AI alignment / computational social science and computer vision (media forensics / deepfake detection). The two cataloged 2026 papers sit on opposite sides of this divide, so current evidence shows two parallel tracks rather than a single concentrated line; more output would be needed to see whether one direction becomes dominant.
Collaboration54/100
The collaboration axis scores 54, mid-range. The snapshot records 14 distinct coauthors, 12 of them recurring partners, including Keqing Zhang, Jingyu Chen, Nai Ding, Bing Li, and Weiming Hu. Both cataloged papers list Lang as a middle author without the corresponding role, pointing to contributions within larger teams rather than solo-led projects; whether the same collaborators span both papers is not recorded in the current data.
Activity—
Year-by-year activity data (publication cadence and recency of output) is not available, so the axis is unscored. It would measure papers per year and the interval since the latest output; only the two 2026 preprints are currently dated.
Reproducibility78/100
The reproducibility axis scores 78, the mean across 2 reviewed papers. The deepfake-detection paper (UCF-Net) is rated high reproducibility readiness, with explicit benchmark numbers — 95.33 in-domain and 92.15 cross-domain mAUC on a 4M-image benchmark; the LLM-values paper is rated medium, with Evolve noting a circularity risk from LLM-generated item labels. Together this suggests above-average methodological transparency for an early-stage record.
Research summary
Congyan Lang is a researcher at Beijing Jiaotong University whose cataloged work spans two parallel tracks: AI alignment with a computational social science orientation, and computer vision for media forensics. Both entries in the Evolve catalog are 2026 arXiv preprints listing Lang as a middle author rather than corresponding author. The first, “Do LLMs Have Values?”, introduces a quantitative analysis and alignment framework for values in large language models, grounded in Schwartz value theory and the World Values Survey; Evolve's review calls the scale impressive but cautions that LLM-based item labels risk circularity. The second, “Harnessing CLIP and DINO”, presents UCF-Net, an uncertainty-aware cascaded fusion network that tops a 4M-image deepfake benchmark with 95.33 in-domain and 92.15 cross-domain mAUC, though zero-shot transfer remains weak. The bibliometric snapshot records 14 distinct coauthors, an even 50/50 field split, a mid-range collaboration score (54), and a strong reproducibility score (78) computed from two reviewed papers. With output, impact, and activity data still missing, this is an early-stage, dual-field profile built strictly from the reviewed papers and the linked ORCID record.
Collaboration pattern
The snapshot records 14 distinct coauthors and 12 recurring partners, among them Keqing Zhang, Jingyu Chen, Yufan Liu, Yongqiang Zhu, Nai Ding, Lai Jiang, Bing Li, Weiming Hu, Xuechao Zou, Yi Zhou, Kai Li, and Shun Zhang. With a collaboration score of 54, Lang sits mid-range: both cataloged papers are middle-author, non-corresponding contributions, and the per-paper team rosters are not separated in the current data, so cluster structure across the two fields remains unverified.
Research keywords
Highlights
Quantifying values in LLMs with the PEC framework
UCF-Net tops a 4M-image deepfake benchmark
Timeline
- 2026Co-authored arXiv preprint “Harnessing CLIP and DINO” introduces UCF-Net, an uncertainty-aware cascaded fusion network for generalizable deepfake image detection.
- 2026Co-authored arXiv preprint “Do LLMs Have Values?” introduces a quantitative analysis and alignment framework for values in large language models.
- 2026The two 2026 preprints split evenly across the researcher's two fields, with Evolve rating reproducibility readiness high for the deepfake work and medium for the values framework.
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 figures come from a limited public bibliometric snapshot; output, impact, and activity axes lack underlying data and are not scored.
- Both cataloged papers are 2026 arXiv preprints; peer-review status and journal or conference acceptance are not reflected in this data.
- The reproducibility score (78) aggregates Evolve reviews of only 2 papers, so it should be read as directional rather than definitive.
- This profile is name-resolved from public bibliographic data linked to ORCID 0000-0001-6059-7943; disambiguation relies on that identifier.