Georgia Tech researcher spanning trustworthy ML certification, physical world models, and quantum machine learning.
4
papers here
3
reviewed by Evolve
1
first-author
0
corresponding
—
works · OpenAlex
—
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-17) 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—
No output metrics exist in this snapshot: publication counts, venues, and yearly volume are absent, and only the three Evolve-reviewed 2026 preprints are on record, so overall research productivity cannot yet be scored.
Impact—
Impact cannot be measured yet: this snapshot contains no citation counts, h-index, or venue acceptance data. This axis would track citation uptake and scholarly influence of the reviewed papers once such records become available.
Focus0/100
The focus score of 0 derives from an evenly split field distribution: each of three lines carries one-third weight — quantum machine learning (Bayesian quantum tomography), machine learning for learned physical simulation, and machine learning for trustworthy ML with statistical guarantees. No specialization dominates the catalog; the profile currently favors breadth across distinct methodological communities over concentration in a single area.
Collaboration48/100
A collaboration score of 44 rests on 8 distinct coauthors across the three reviewed preprints. Yufeng Wang is the most frequent collaborator, with Haibin Ling, Lu Wei, and Michael Chaykowsky also recurring. This points to a moderate-sized, stable working circle rather than a large distributed network — consistent with mid-range collaboration intensity rather than either solo work or extensive multi-team partnerships.
Activity—
Activity data is insufficient: the snapshot covers only three 2026 arXiv preprints, with no multi-year publication cadence. This axis would measure research tempo — output frequency and continuity over time — and remains unscored here.
Reproducibility85/100
Reproducibility is the strongest measured axis: a mean readiness score of 85 across all three reviewed papers (reproPapers = 3). Two preprints are rated high readiness — the selective-prediction certification work and the conservation-based world-model study — while GRPO-QPS for quantum posterior sampling is rated medium. Evolve's review suggests careful methodological and empirical documentation across this catalog.
Research summary
Parivesh Priye is a researcher at Georgia Institute of Technology whose reviewed work divides almost evenly across three lines: trustworthy machine learning with statistical guarantees, learned physical simulation, and quantum machine learning for Bayesian quantum tomography. The Evolve catalog holds three 2026 arXiv preprints. The first-authored 'Available Guardrails: Certifying Selective Prediction across ML Systems' proves an exact binomial availability identity and applies dynamic-programming partition planning, Clopper–Pearson intervals, and multiple-hypothesis control to certify selective prediction; Evolve rates it rigorous but with modest gains over random baselines. 'Conservation Buys Stability and Factoring Buys Counterfactuals in Physical World Models' argues that symplectic conservation underwrites stability while factored coupling enables counterfactual generalization in Hamiltonian-style learned simulators — strong synthetic evidence, though the inversion result is partly by construction. 'GRPO-QPS' trains Metropolis-corrected MCMC proposals with GRPO so they preserve the quantum-tomography posterior, with controls showing the payoff concentrates on multimodal targets. Reproducibility is the clearest measured signal: a mean readiness score of 85 across all three papers, two rated high and one medium. Collaboration is compact — 8 distinct coauthors, most frequently Yufeng Wang, plus Haibin Ling, Lu Wei, and Michael Chaykowsky. Output, impact, and activity metrics are not yet available, so citation-based judgment should wait.
Collaboration pattern
Co-authorship in the reviewed catalog is compact: 8 distinct collaborators across three 2026 preprints, with a collaboration score of 44 (mid-range). Yufeng Wang appears most frequently, followed by recurring collaborations with Haibin Ling, Lu Wei, and Michael Chaykowsky. The pattern suggests a small, stable working group whose interests span trustworthy ML, physical simulation, and quantum topics, rather than a broad multi-institution network; all cataloged papers are arXiv preprints.
Research keywords
Highlights
Exact statistical certification for selective prediction
Conservation yields stability; factoring yields counterfactuals
RL-trained proposals for quantum posterior sampling
Timeline
- 2026Posts 'GRPO-QPS: Target-Preserving Reinforcement Learning for Quantum Posterior Sampling' (arXiv 2609.14711), the earliest of the three cataloged 2026 preprints.
- 2026Posts 'Conservation Buys Stability and Factoring Buys Counterfactuals in Physical World Models' (arXiv 2609.19674), linking symplectic conservation to stability and factored coupling to counterfactuals.
- 2026First-authors 'Available Guardrails: Certifying Selective Prediction across ML Systems' (arXiv 2609.22048), proving an exact binomial availability identity for selective prediction.
- 2026All three 2026 preprints enter the Evolve catalog with a mean reproducibility readiness of 85, spanning trustworthy ML, learned physical simulation, and quantum machine learning.
Papers in the Evolve catalog
- first author
Available Guardrails: Certifying Selective Prediction across ML Systems
Conservation Buys Stability and Factoring Buys Counterfactuals in Physical World Models
GRPO-QPS: Target-Preserving Reinforcement Learning for Quantum Posterior Sampling
VoiceLongMemEval: Do Assistants Remember How You Sounded?
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.
- Scores are computed from a public bibliometric snapshot only; no citation counts, h-index, or venue records are included, so the output, impact, and activity axes remain unscored.
- Evidence is limited to three Evolve-reviewed 2026 arXiv preprints; earlier, non-arXiv, or non-reviewed work is outside this profile's scope.
- No ORCID or country is provided; the profile is name-resolved from public bibliographic data and same-name attribution cannot be fully excluded.
- Field labels and verdicts reflect Evolve's review of the listed preprints; peer-reviewed publication status is unknown.