Shuaijun Liu
The Hong Kong University of Science and Technology (Guangzhou)
HKUST (Guangzhou) researcher linking latent world models for robot learning with state-space and Bayesian filtering theory.
2
papers here
2
reviewed by Evolve
1
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—
Not yet scored: no publication-count basis is available in this snapshot, so output — the volume, venue mix and cadence of papers over time — cannot be measured from current data.
Impact—
Not yet scored: citation counts and downstream adoption are absent from the snapshot, so impact — how strongly later research and practice build on this work — remains unmeasured.
Focus0/100
Scores 0/100, mirroring an exact 50/50 field split between machine learning theory (sequential state-space models and Bayesian filtering) and robot learning / world models: one of the two cataloged 2026 papers sits in each field. The profile is genuinely dual-track rather than concentrated; with only two scored entries, this reads as early-stage breadth rather than established diffusion across topics.
Collaboration56/100
Scores 56/100. The snapshot records 16 distinct co-authors, with Ningxin Su and Chengyu Wu each appearing twice in the frequency list alongside Qifu Wen, Zihan Zhou, Xi Zeng, Feiyang You, Chenglong Zhang, Shuyang Hao and Xi Lin. That circle is real but still small, and the mid-range score matches it: first-author leadership on the D-JEPA robotics project combined with a middle-author role on the theory paper.
Activity—
Not yet scored: per-year publication counts are missing, so activity — sustained research tempo over time — cannot be assessed; both cataloged papers carry 2026 dates only.
Reproducibility78/100
Scores 78/100, the mean across 2 reproducibility-scored papers. The Bayesian-filtering theory paper rates high: rigorous proofs with an honestly bounded scope. D-JEPA rates medium: gains are consistent in simulation and on a real robot, but cohort selection, a tiny driving set and no released code limit independent replication.
Research summary
Shuaijun Liu is a researcher affiliated with The Hong Kong University of Science and Technology (Guangzhou) whose reviewed work spans two complementary tracks: machine learning theory for sequential state-space models and Bayesian filtering, and robot learning built on latent world models. The Evolve catalog holds two 2026 arXiv papers. As first author, Liu proposed D-JEPA, a decision-aligned latent world model that re-ranks latent-world-model candidate futures using a bounded set transformer to close a decision-local gap; Evolve's verdict notes consistent simulation and real-robot gains, while flagging cohort selection, a tiny driving set and the absence of released code. The second paper, a middle-author contribution, proves that in Gaussian hidden Markov models fixed-K diverging Bayesian updates and radial updates can coexist with vanishing predictive KL divergence — a result Evolve describes as rigorously proved with an honestly bounded scope and high reproducibility readiness. Quantitatively, the radar shows a reproducibility score of 78 averaged over two scored papers, collaboration at 56 based on 16 distinct co-authors, and a focus score of 0, reflecting an even 50/50 field split. Output, impact and activity axes lack sufficient data and are left unscored. With only two reviewed preprints, this is an early-stage profile: promising breadth across theory and robot learning, with depth still to be established.
Collaboration pattern
Sixteen distinct co-authors appear across the two cataloged 2026 arXiv papers; Ningxin Su and Chengyu Wu each appear twice in the frequency list, alongside Qifu Wen, Zihan Zhou, Xi Zeng, Feiyang You, Chenglong Zhang, Shuyang Hao and Xi Lin. First-authorship of D-JEPA suggests leadership in a robotics-oriented team, while the theory paper is a middle-author contribution. The frequency list also contains an entry matching the researcher's own name — a likely entity-resolution artifact.
Research keywords
Highlights
First-authored D-JEPA closes a decision-local gap in latent world models
Theory result: good predictors can still diverge as predictive KL vanishes
Timeline
- 2026Posts D-JEPA to arXiv as first author — a decision-aligned latent world model that re-ranks candidate futures with a bounded set transformer.
- 2026Co-authors the theory paper proving fixed-K diverging Bayesian and radial updates coexist with vanishing predictive KL in Gaussian hidden Markov models.
- 2026Both papers enter the Evolve review catalog with reproducibility readiness of medium and high, yielding a mean reproducibility score of 78.
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 scores are derived from public bibliometric data (OpenAlex-based); no citation counts, h-index or venue-tier metrics were available, so output, impact and activity are unscored.
- Only two reviewed papers, both 2026 arXiv preprints, underpin this profile; peer-reviewed publication status is not recorded in the snapshot.
- The profile is name-resolved from public bibliographic data; the frequent-coauthor list contains an entry matching the researcher's own name and duplicates (Ningxin Su, Chengyu Wu), indicating possible entity-resolution noise.
- Country and ORCID are empty in the source data, so details beyond the stated institution remain unverified.