ML researcher bridging optimization theory, learned physical simulation, and certified trustworthy machine learning
4
papers here
4
reviewed by Evolve
0
first-author
0
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—
Output metrics are not yet available for this profile: no publication counts, venue list, or productivity trend exist in the current snapshot. This axis would measure total research output and its growth over time, and is deliberately left unscored.
Impact—
Impact data such as citation counts, h-index, or downstream adoption is absent from this snapshot. The axis would measure how widely the three 2026 preprints are cited and used; given their recency, it remains unscored rather than estimated.
Focus0/100
The focus score is 0, reflecting an evenly spread field distribution: optimization theory, learned physical simulation, and trustworthy ML with statistical guarantees each account for exactly one third (0.333) of the cataloged output. All three 2026 papers fall in distinct subfields, so no single direction yet dominates. This breadth supports cross-area method transfer but means a consolidated research identity has not emerged from the current three-paper record.
Collaboration51/100
The collaboration score is 44 with 8 distinct coauthors observed, a modest-sized network. The frequent-coauthor list includes Yufeng Wang, Parivesh Priye, Yihan Zhang, Rohit Dhaipule, Chiu C Tan, Hongchang Gao, Lu Wei, and Michael Chaykowsky. Authorship positions vary — middle on two papers and last on the physical-simulation paper — and none is flagged corresponding, suggesting a substantive contributor role within small teams rather than a large-lab lead pattern.
Activity—
Activity data — publication cadence, preprint frequency, and year-over-year rhythm — is not in the current snapshot. All three cataloged papers carry 2026 dates, which is too narrow a window to score pace. The axis would measure research tempo over time.
Reproducibility78/100
The reproducibility score is 85, computed as the mean over 3 papers. Two papers are rated high readiness — the selective-prediction certification work and the physical world models paper — while the federated bilevel optimization paper is rated medium, consistent with Evolve's note that its experiments are small-scale and single-run. Theory-heavy contributions with clearly stated guarantees and synthetic setups appear to drive this above-average score.
Research summary
Haibin Ling is a machine learning researcher whose recent work spans three near-equal threads — optimization theory, learned physical simulation, and trustworthy machine learning with statistical guarantees — each accounting for about a third of the field distribution in this profile. The cataloged record comprises three 2026 arXiv preprints. In trustworthy ML, Ling co-authored 'Available Guardrails', which proposes an exact binomial availability identity and dynamic-programming partition planning for certifying selective prediction across ML systems; Evolve's review calls the work rigorous while noting modest gains over random baselines. In learned physical simulation, Ling authored last 'Conservation Buys Stability and Factoring Buys Counterfactuals', arguing that symplectic conservation yields stability and factored coupling enables counterfactual generalization, with strong synthetic evidence though partly by-construction inversion. In optimization, Ling co-authored the first fully first-order federated bilevel method with an O(κ¹⁰/(Nε⁵)) linear-speedup guarantee, though experiments are small-scale and single-run. The profile shows 8 distinct coauthors and a collaboration score of 44, plus a reproducibility score of 85 averaged over three papers, with two rated high readiness and one medium. Output, impact, and activity metrics are not yet available in the data scope, so this profile should be read as an early-stage, name-resolved record.
Collaboration pattern
Collaboration occurs in small teams: 8 distinct coauthors appear across the three 2026 preprints, and the collaboration score is 44. Recurring partners include Yufeng Wang, Parivesh Priye, Yihan Zhang, Rohit Dhaipule, Chiu C Tan, Hongchang Gao, Lu Wei, and Michael Chaykowsky. Authorship is mixed — middle positions on the certification and optimization papers, last position on the physical-simulation paper — and no cataloged paper flags corresponding authorship.
Research keywords
Highlights
Exact availability identity certifies selective prediction across ML systems
Conservation and factorization as principles for physical world models
First fully first-order federated bilevel optimization method
Timeline
- 2026Co-authored 'Available Guardrails: Certifying Selective Prediction across ML Systems' on arXiv (middle author).
- 2026Last-authored 'Conservation Buys Stability and Factoring Buys Counterfactuals in Physical World Models' on arXiv.
- 2026Co-authored 'Federated stochastic bilevel optimization with fully first-order gradients' on arXiv (middle author).
Papers in the Evolve catalog
Available Guardrails: Certifying Selective Prediction across ML Systems
Conservation Buys Stability and Factoring Buys Counterfactuals in Physical World Models
Federated stochastic bilevel optimization with fully first-order gradients
GRPO-QPS: Target-Preserving Reinforcement Learning for Quantum Posterior Sampling
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 (collaboration 44, focus 0, reproducibility 85) are computed from public bibliometric data; output, impact, and activity aggregates are not available in this snapshot.
- No institution, country, or ORCID is attached to this record; the profile is name-resolved from public bibliographic data and may share a name with other researchers.
- All three cataloged papers are 2026 arXiv preprints; no peer-reviewed venue, citation counts, or funding information is in scope.
- The federated bilevel optimization paper carries no source keywords; its related keywords here are inferred from the title and Evolve's review notes.