Boston University researcher linking state-space model theory with decision-aligned world models for robot learning
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—
Output metrics are not yet available: the underlying data does not quantify publication counts or yearly productivity. This axis would measure total and per-year paper output across the career, beyond the two reviewed 2026 arXiv papers.
Impact—
Impact data is unavailable: citation counts, h-index, and downstream adoption are not in the current snapshot. This axis would measure how often the researcher's work is cited and built upon, which cannot yet be scored.
Focus0/100
The computed focus score is 0/100, driven by an even 50/50 field distribution: half of the reviewed output sits in machine learning theory (sequential state-space models and Bayesian filtering) and half in robot learning / world models. Across only two 2026 papers — a first-author theory paper and a middle-author robotics paper — this spread reads as genuine dual-track work rather than an emerging concentration; more papers would clarify whether one direction consolidates.
Collaboration56/100
Collaboration scores 56/100 on 16 recorded distinct coauthors. The frequent-coauthor list repeats Shuaijun Liu, Ningxin Su, and Chengyu Wu, suggesting a small recurring core alongside a wider periphery. Authorship roles differ by track: first author on the theory paper but middle author, non-corresponding, on the D-JEPA robotics paper — consistent with contributing within a larger applied team while leading theory work independently.
Activity—
Activity metrics are not available: yearly publication cadence and recent-output trends are not computed in this snapshot. The axis would measure publishing consistency over time; two 2026 arXiv papers alone cannot establish an activity pattern.
Reproducibility78/100
Reproducibility is the strongest computed axis at 78/100, a mean taken over 2 reviewed papers. The first-author theory paper rates high — rigorous proofs with honestly bounded scope need no code release — while D-JEPA rates medium, with consistent simulation and real-robot gains but cohort selection concerns, a tiny driving set, and no code. The gap between the two is itself informative.
Research summary
Qifu Wen is a researcher at Boston University whose published work spans two complementary halves of modern machine learning. On the theoretical side, Wen is first author of 'How Wrong Can a Good Predictor Be? Diverging Updates with Vanishing Predictive KL' (2026, arXiv), which proves that fixed-K diverging Bayesian and radial updates can coexist with vanishing predictive KL divergence in Gaussian hidden Markov models — a result relevant to selective state-space models such as Mamba and to softmax decoder sensitivity. The Evolve review rates its reproducibility readiness as high, citing rigorous proofs and honestly bounded scope. On the empirical side, Wen contributed as a middle author to 'D-JEPA: A Decision-Aligned Latent World Model' (2026), which re-ranks latent-world-model candidate futures through a bounded set transformer to close a decision-local gap; Evolve notes consistent simulation and real-robot gains, while flagging cohort selection, a small driving set, and no released code, with medium reproducibility readiness. The bibliometric snapshot records 16 distinct coauthors and an even 50/50 split between machine learning theory (sequential state-space models and Bayesian filtering) and robot learning / world models, with a mean reproducibility score of 78 across the two reviewed papers. Output, impact, and activity metrics are not yet available, so this profile remains early-stage and rests on the two reviewed papers.
Collaboration pattern
The data records 16 distinct coauthors, with Shuaijun Liu, Ningxin Su, and Chengyu Wu each appearing twice in the frequent-coauthor list — the clearest signal of a recurring core team. The pattern suggests fluid, project-based collaboration: Wen leads theory work as first author while slotting into a larger applied roster on D-JEPA. Coauthor affiliations are not in the snapshot, so cross-institution ties cannot be confirmed.
Research keywords
Highlights
First-author theory result: divergent updates with vanishing predictive KL
Decision-aligned latent world model with consistent sim-to-robot gains
Dual-track portfolio: selective state-space theory meets world-model planning
The two 2026 papers split evenly between sequential state-space / Bayesian filtering theory and robot-learning world models; shared threads — predictive KL, softmax decoder sensitivity, and model-based planning — link Mamba-style selective SSMs to decision-time model use.
Timeline
- 2026Publishes first-author theory paper 'How Wrong Can a Good Predictor Be? Diverging Updates with Vanishing Predictive KL' on arXiv.
- 2026Contributes as middle author to 'D-JEPA: A Decision-Aligned Latent World Model', a robot-learning world model paper on arXiv.
- 2026Both catalog papers appear in 2026, pairing theory and robotics within the same year; no earlier career timeline is present in the data.
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 inputs (coauthor count, field distribution, focus and reproducibility scores) come from OpenAlex-derived data and cover only the Evolve-reviewed subset.
- Output, impact, and activity axes are missing from the radar; no citation counts or yearly publication totals were available.
- Only two 2026 arXiv papers are reviewed; earlier work or other venues is outside this profile's evidence base.
- The profile is name-resolved from public bibliographic data (ORCID 0009-0009-8458-144X); no biographical details beyond affiliation are verified.