Yufeng Wang

Stony Brook University

Unclaimed · AI-structured profileOpenAlex

Stony Brook ML researcher: certified selective prediction and physics-grounded learned world models

papers here

reviewed by Evolve

first-author

corresponding

—

works · OpenAlex

—

h-index · OpenAlex

Bibliometric figures come from OpenAlex (2026-10-15) 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: only two 2026 arXiv papers are cataloged for this name, so publication volume cannot be computed. This axis would measure total papers, venue spread, and yearly output once a fuller bibliometric record is linked.

Impact—

Not yet scored: no citation counts, h-index, or downstream adoption data are present in the input. This axis would measure citations and uptake of the certified-prediction and world-model papers once citation data are linked.

Focus0/100

Scores 0 on focus, reflecting an observed field distribution split exactly 50/50 between learned physical simulation and trustworthy ML with statistical guarantees. Within this small sample the two 2026 papers sit in distinct subfields — one certifying selective prediction with exact binomial identities and dynamic programming, the other on Hamiltonian, symplectically integrated world models — so no dominant single topic emerges yet from the cataloged record.

Collaboration39/100

Scores 39 on collaboration, derived from 6 distinct coauthors across the two cataloged papers. The recurring partner set includes Parivesh Priye, Haibin Ling, Lu Wei, and Michael Chaykowsky, suggesting a small, repeat collaborator circle rather than a broad network. Authorship roles differ across the record: first author on the physical world models paper and middle author on the selective prediction certification paper.

Activity—

Not yet scored: yearly publication cadence cannot be derived from two same-year (2026) records. This axis would measure recent publication rate and consistency over time once a longer publication history is available.

Reproducibility100/100

Scores 100, the maximum, from a reproducibility mean of 100 across 2 papers — both cataloged works carry a high reproducibility readiness rating. This reflects Evolve's assessment that the exact binomial availability identity with Clopper–Pearson intervals and the symplectic, energy-conserving world-model experiments are concretely checkable, though the certification paper's gains over random baselines are modest and the world-model inversion is partly by construction.

Research summary

Yufeng Wang is a machine learning researcher at Stony Brook University whose cataloged work splits evenly, 50/50 by field distribution, between learned physical simulation and trustworthy machine learning with statistical guarantees. Evolve's catalog includes two 2026 arXiv papers. The first-authored 'Conservation Buys Stability and Factoring Buys Counterfactuals in Physical World Models' argues that stability in learned world models requires symplectic conservation while counterfactual generalization requires factored coupling; Evolve judged the synthetic evidence strong while noting that inversion is partly by construction. The middle-authored 'Available Guardrails: Certifying Selective Prediction across ML Systems' proposes an exact binomial availability identity and dynamic-programming partition planning for selective prediction, built on Clopper–Pearson intervals, multiple hypothesis testing, and conformal risk control; it is rated rigorous, though gains over random baselines are modest. Reproducibility is the standout signal: a reproducibility mean of 100 across 2 papers, both rated high readiness. Collaboration is compact — 6 distinct coauthors, a collaboration score of 39, and recurring partners including Parivesh Priye, Haibin Ling, Lu Wei, and Michael Chaykowsky. Focus scores 0, an honest reflection of the even two-field split in this small sample. Output, impact, and activity cannot be scored from the current record and remain unscored pending fuller bibliometric data.

Collaboration pattern

The observed collaboration footprint is compact: 6 distinct coauthors across two 2026 papers, giving a collaboration score of 39. Recurring names include Parivesh Priye, Haibin Ling, Lu Wei, and Michael Chaykowsky; the raw record lists Haibin Ling twice and includes the researcher's own name, likely disambiguation artifacts. Authorship spans first author on the physical world models paper and middle author on the certification work, indicating both leading and supporting roles.

Research keywords

selective predictionstatistical certificationClopper–Pearson intervalmultiple hypothesis testingconformal risk controldynamic programmingHamiltonian neural networkssymplectic integrationcounterfactual generalizationphysical world modelslearned simulatorsenergy conservation

Highlights

First-authored claim: conservation buys stability, factoring buys counterfactuals
In the 2026 arXiv paper 'Conservation Buys Stability and Factoring Buys Counterfactuals in Physical World Models' (first author), Wang argues that learned world models need symplectic conservation for stability and factored coupling for counterfactual generalization. Evolve finds the synthetic evidence strong but notes the inversion result is partly by construction; reproducibility readiness is rated high.
Certifying selective prediction with exact binomial guarantees
The 2026 arXiv paper 'Available Guardrails: Certifying Selective Prediction across ML Systems' (middle author) introduces an exact binomial availability identity and dynamic-programming partition planning, drawing on Clopper–Pearson intervals, multiple hypothesis testing, and conformal risk control. Evolve calls the work rigorous while flagging that gains over random baselines are modest; reproducibility readiness is high.
Full-score reproducibility across the cataloged record
Both cataloged papers carry a high reproducibility readiness rating, giving a reproducibility mean of 100 over 2 papers — the strongest quantified signal in this profile.

Timeline

2026Posts the first-authored arXiv paper on conservation and factoring in physical world models
2026Co-authors the arXiv paper certifying selective prediction across ML systems (middle author)
2026Both papers enter the Evolve catalog with high reproducibility readiness (mean 100 over 2 papers)

Papers in the Evolve catalog

2026-09-18
Available Guardrails: Certifying Selective Prediction across ML Systems
2026-09-18first author
Conservation Buys Stability and Factoring Buys Counterfactuals in Physical World Models

Frequent co-authors

Parivesh Priye×2 Haibin Ling×2 Lu Wei×1 Yufeng Wang×1 Haibin Ling×1 Michael Chaykowsky×1

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.

Radar dimensions are computed from a thin bibliometric sample: only two 2026 arXiv papers, so output, impact, and activity are marked missing rather than estimated.
No citation counts, h-index, venue acceptance, funding, or career-history data were provided; none are inferred.
'Yufeng Wang' is a common name; this profile is name-resolved from public bibliographic data and may conflate or omit same-name authors.
The frequent-coauthor list contains 'Haibin Ling' twice and the researcher's own name, likely disambiguation artifacts in the source record.