ML researcher working across AI-alignment failure detection and multimodal graph foundation models
2
papers here
0
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
—
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-09-27) 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—
Not yet scored: this snapshot contains no publication-volume data beyond the two reviewed papers. This axis would measure total output, venue distribution, and yearly production across the full publication record.
Impact—
Not yet scored: no citation counts or h-index data are present. This axis would measure citation-based influence — total citations, h-index, and per-paper impact — once a fuller bibliometric record is linked.
Focus—
Not yet scored: with only two reviewed papers, topical concentration cannot be computed. This axis would measure how consistently research themes and fields are maintained across the full publication history.
Collaboration54/100
Scored at 54/100 on the collaboration axis, derived from 14 distinct coauthors in the bibliometric snapshot. Twelve collaborators recur, including Xunkai Li, Rong-Hua Li, Guoren Wang, Gelei Deng, Yuekang Li, and Simin Chen. Both cataloged 2026 papers place Yi Liu in a middle author position rather than first or corresponding, a pattern consistent with substantive contribution inside larger, stable teams rather than solo or lead-author work.
Activity—
Not yet scored: the snapshot provides publication years only for the two reviewed entries. This axis would measure publication cadence and research continuity over time across the complete record.
Reproducibility—
Not yet scored at the axis level. This axis would measure code and data release, seed and variance reporting, and artifact completeness; at paper level, Evolve flags the Jev paper high and ICE medium on reproducibility readiness.
Research summary
Yi Liu appears in the Evolve catalog as a coauthor on two 2026 arXiv preprints spanning two distinct areas of machine learning. The first, 'Just Ask Jev' (2026), benchmarks a reinforcement-learning-from-calibrated-decisions (RLCD) model, Jev, as a zero-shot detector of AI alignment failures, reporting a median AUROC of 0.886; Evolve rates its methods strong and its reproducibility readiness high, while noting the evaluation scope is limited to a single model vendor. The second, 'ICE' (2026), proposes task-aligned Clifford latent fields — a Cl(3) formulation — for multimodal graph foundation models and claims first place across 30 multimodal graph tasks; Evolve judges the results plausible but contingent on 8-seed means, quoted baselines, and absent code. In both papers Liu holds a middle author position rather than first or corresponding. The bibliometric snapshot records 14 distinct coauthors and a collaboration dimension score of 54, with a recurring collaborator group including Xunkai Li, Rong-Hua Li, Guoren Wang, Gelei Deng, Yuekang Li, and Simin Chen. Citation counts, publication volume, institution, and ORCID are not available in this snapshot, so the output, impact, focus, and activity axes remain unscored, and this profile should be read as name-resolved from public bibliographic data for a relatively common name.
Collaboration pattern
The snapshot records 14 distinct coauthors and a collaboration score of 54, with twelve recurring collaborators: Xunkai Li, Xu Wang, Yinlin Zhu, Xiong Yongfu, Rong-Hua Li, Guoren Wang, Ruoqi Guo, Gelei Deng, Yuekang Li, Lida Zhao, Yutao Wu, and Simin Chen. Both cataloged papers place Yi Liu in a middle author position, suggesting a contributor role within established teams; the snapshot does not attribute individual coauthors to specific papers.
Research keywords
Highlights
Calibrated RLCD signals as zero-shot alignment-failure detection
Cl(3) Clifford latent fields for multimodal graph foundation models
Middle-author contributor embedded in a stable collaborator network
Both cataloged 2026 papers list Yi Liu as a middle author rather than first or corresponding author, and the snapshot records 14 distinct coauthors with twelve recurring collaborators — a collaborative, team-embedded working pattern rather than a solo, lead-author profile.
Timeline
- 2026Co-authored 'Just Ask Jev': reinforcement learning for calibrated decisions (RLCD) benchmarked as a zero-shot detector of AI alignment failures, with a reported median AUROC of 0.886.
- 2026Co-authored 'ICE', proposing task-aligned Cl(3) Clifford latent fields for multimodal graph foundation models, with claimed first-place results across 30 multimodal graph tasks.
- 2026Both preprints appear on arXiv with Yi Liu in a middle author position; Evolve rates reproducibility readiness high for the Jev paper and medium for ICE.
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.
- Profile is name-resolved from public bibliographic data: 'Yi Liu' is a common name, and institution, country, and ORCID are not provided in this snapshot.
- Bibliometric dimensions come from an OpenAlex-derived snapshot; only the collaboration axis had sufficient data (score 54, based on 14 distinct coauthors).
- Coverage is limited to the two arXiv papers reviewed by Evolve; citation counts, venue history, and full publication output fall outside this data scope.
- Frequent coauthors are aggregated across the snapshot and are not attributed to individual papers.