David Yallup

University of Cambridge · United Kingdom

Unclaimed · AI-structured profileOpenAlex

David Yallup — Bayesian computation researcher at Cambridge building neural and gradient-based upgrades to nested sampling

papers here

reviewed by Evolve

first-author

corresponding

—

works · OpenAlex

h-index · OpenAlex

Bibliometric figures come from OpenAlex (2026-09-27) 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 volume cannot be scored yet: the raw record reports h-index 0 and i10-index 0 with no citation counts, and countsByYear contains only 2026. What this axis would measure is the number of publications and their volume trend over time; until a longer publication window and indexed counts exist, no estimate is made.

Impact—

Impact is missing from the underlying data: h-index and i10-index are both 0 and no citation counts are reported, so this axis is unscored. It would measure citations, citation decile, and uptake of the two reviewed 2026 preprints; as arXiv-first work, indexed citations have not accumulated and no estimate is offered.

Focus100/100

Scores 100, reflecting an exceptionally concentrated research agenda. All 14 recent-three-year works sit in a single field distribution entry (Bayesian computation / machine learning, sampling methods), and topic counts stack on Gaussian processes and Bayesian inference (7) and Markov chains and Monte Carlo methods (6), with generative adversarial networks and image synthesis (4) as the main secondary cluster.

Collaboration14/100

Scores 14, the weakest measured axis: the raw record lists only 1 distinct coauthor, and the sole frequent collaborator is Will Handley. Both reviewed 2026 papers are first-authored (one also corresponding), suggesting a tight two-person working pattern rather than a large distributed team; this is typical of early-stage profiles and could broaden with new projects.

Activity79/100

Scores 79 on the strength of recent output volume: recent3yWorks is 14 and countsByYear attributes all 14 works to 2026. Every catalogued paper is from 2026, indicating a dense, current burst of method-building in sampling algorithms rather than a long publication history; the record is snapshot-limited to a single year.

Reproducibility100/100

Scores 100, the maximum, based on reproMean 100 across reproPapers equal to 1. It aligns with Evolve's review verdicts: Neural Transport Nested Sampling is rated high reproducibility readiness, and Quenched Ensemble Sampling is likewise high and described as convincingly well-caveated. Note the small sample — one scored paper — so this top score should be read narrowly.

Research summary

David Yallup is a researcher at the University of Cambridge working at the intersection of Bayesian computation and machine learning, with a sharp focus on sampling methods. The bibliometric record (OpenAlex) shows a highly concentrated output profile: all 14 works in the recent three-year window fall in 2026, and topic assignments cluster in Gaussian processes and Bayesian inference (7), Markov chain and Monte Carlo methods (6), and generative adversarial networks and image synthesis (4), with smaller footprints in fault detection and control systems (2) and quantum many-body systems (2). Evolve has reviewed two of his 2026 preprints, both with him as first author: Neural Transport Nested Sampling, which adds a flow-matching drift to nested sampling's replacement step and achieves strong Lennard-Jones cluster accuracy with calibrated partition-function estimates, and Quenched Ensemble Sampling, which replaces nested sampling's hard energy wall with repulsive potentials so gradient-based sequential Monte Carlo can cross first-order phase transitions. Both papers carry a high reproducibility-readiness rating from Evolve, consistent with a reproducibility score of 100 in the underlying data. The collaboration footprint is currently narrow — one distinct coauthor in the raw record, with Will Handley the frequent collaborator — and citation-based output and impact metrics (h-index 0, i10-index 0) are not yet informative, so this should be read as an early-stage, activity- and method-focused profile.

Collaboration pattern

The collaboration record is deliberately small: the raw data lists 1 distinct coauthor, and the only frequent collaborator is Will Handley, also at Cambridge. Both reviewed 2026 papers are first-authored (one additionally corresponding), which points to a close, sustained two-person collaboration on nested-sampling methodology rather than wide network-based work. Expect collaborative breadth to grow as the gradient-based sampling programme matures.

Research keywords

nested samplingsequential Monte Carlomarginal likelihood estimationflow matchingMALA / gradient-based MCMCBoltzmann samplingpartition function estimationfirst-order phase transitionsLennard-Jones clustersMetropolis-Hastings

Highlights

Neural Transport Nested Sampling
First-authored 2026 preprint adding a flow-matching drift to nested sampling's replacement step. Evolve's verdict: strong Lennard-Jones (LJ-55) accuracy and calibrated partition function Z(β), though headline numbers are median-seed picks; reproducibility readiness rated high.
Quenched Ensemble Sampling
First-authored and corresponding 2026 preprint replacing nested sampling's hard energy wall with repulsive potentials, letting gradient-based sequential Monte Carlo cross first-order phase transitions. Evolve judges the approach convincing and well-caveated, with high reproducibility readiness.
A coherent neural-transport programme for nested sampling
Taken together, the two 2026 papers show a single research line: injecting learned transport and gradient information into nested sampling, first in the replacement step, then in the constraint itself via repulsive potentials. Both are high-reproducibility and grounded in MALA / gradient-based MCMC machinery.

Timeline

2026Publishes Neural Transport Nested Sampling (arXiv), introducing a flow-matching drift into nested sampling's replacement step with strong LJ-55 results.
2026Publishes Quenched Ensemble Sampling (arXiv) as first and corresponding author, replacing the hard energy wall with repulsive potentials for gradient-based SMC.
2026Records 14 works in the recent three-year window, concentrated in Bayesian inference and Monte Carlo topics, with Will Handley as the recurring coauthor.

Papers in the Evolve catalog

2026-09-24first author
Neural Transport Nested Sampling
2026-09-14first authorcorresponding
Quenched Ensemble Sampling

Frequent co-authors

Will Handley×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.

Bibliometric statistics (h-index, i10-index, countsByYear, topics) come from OpenAlex only; citation-based measures may undercount arXiv-first work.
h-index and i10-index are 0 and output/impact dimensions are unscored; this reflects a thin citation record, not an assessed lack of impact.
countsByYear covers only 2026, so multi-year activity trends cannot be computed from this snapshot.
Reproducibility score of 100 is based on a single scored paper (reproPapers = 1) and should be read as a narrow sample.