Eric Moulines

Mohamed bin Zayed University of Artificial Intelligence

Unclaimed · AI-structured profileOpenAlex

MBZUAI researcher at the intersection of conformal prediction theory and optimizer benchmarking for diffusion models.

papers here

reviewed by Evolve

first-author

corresponding

—

works · OpenAlex

—

h-index · OpenAlex

Bibliometric figures come from OpenAlex (2026-10-09) 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—

Unscored: publication counts, venue record, and total authorship volume are not available in this snapshot. This axis would measure cumulative and per-year output across the researcher's full bibliographic record.

Impact—

Unscored: citation counts, h-index, and downstream uptake are absent from the input data. This axis would measure how widely the researcher's methods and results are adopted and cited by later work.

Focus0/100

Focus scores 0/100 because the field distribution is perfectly split: 50% mathematical statistics / machine learning theory (conformal prediction) and 50% machine learning (optimization for diffusion models). Each of the two cataloged 2026 papers occupies one side of this divide, so the profile currently reads as two parallel research lines rather than a single concentrated agenda; the score reflects breadth across these subfields, not inactivity.

Collaboration51/100

The collaboration score is 51/100, built on 12 distinct co-authors in the raw record. The recurring circle includes Rustam Isaev, Anton Conrad, Denis Belomestny, Sergey Samsonov, Arman Bolatov, Egor Shulgin, David Li, Abduragim Shtanchaev, Sebastian U. Stich, Maxim Panov, Peter Richtárik, and Martin Takáč. On both cataloged 2026 papers the researcher sits in a middle, non-corresponding authorship position, indicating contribution within team efforts rather than solo or lead roles.

Activity—

Unscored: per-year publication cadence and recent momentum metrics were not provided. This axis would measure ongoing research output over recent years and whether publishing pace is rising, steady, or slowing.

Reproducibility78/100

Reproducibility averages 78/100 across the 2 papers Evolve has assessed (reproPapers = 2, reproMean = 78). The controlled optimizer benchmark for discrete diffusion is rated high readiness — described as well-run and reproducible — while the conformalized quantile regression theory paper is rated medium, typical of proof-heavy work with fewer executable artifacts. On the evidence available, methodological transparency and benchmark hygiene are above average.

Research summary

Eric Moulines is a researcher at Mohamed bin Zayed University of Artificial Intelligence (MBZUAI) whose Evolve catalog profile is built on two 2026 arXiv papers, in both of which the researcher holds a middle authorship position without corresponding status. The first line of work is mathematical statistics and machine learning theory: a conformalized quantile regression (CQR) paper that derives finite-sample Lp bounds on prediction-interval length and coverage under known covariate shift, paired with matching minimax fixed-score benchmarks; Evolve's verdict calls the theory rigorous, honest, and assumption-bounded. The second line is machine learning systems: described as the first controlled optimizer benchmark for discrete diffusion, comparing AdamW, Muon, and SOAP, which Evolve rates well-run and reproducible while noting text-task budgets of only 3% of published training lengths. These strands split the field distribution evenly (50/50), yielding a focus score of 0, while reproducibility readiness averages 78 across the two assessed papers and collaboration registers 51 on 12 distinct co-authors. Output, impact, and activity metrics are missing from this snapshot, so productivity and citation influence remain unassessed here; conclusions should be read as early-stage and catalog-scoped.

Collaboration pattern

The snapshot records 12 distinct co-authors, with a stable recurring circle: Rustam Isaev, Anton Conrad, Denis Belomestny, Sergey Samsonov, Arman Bolatov, Egor Shulgin, David Li, Abduragim Shtanchaev, Sebastian U. Stich, Maxim Panov, Peter Richtárik, and Martin Takáč. On both cataloged 2026 papers the researcher appears in a middle, non-corresponding authorship position, consistent with a collaborative, theory-heavy working style within team efforts (collaboration score 51/100).

Research keywords

conformal predictionconformalized quantile regressioncovariate shiftconditional coverageminimax lower boundsLe Cam two-point methoddiffusion modelsdiscrete diffusionoptimizer benchmarkAdamWMuonSOAP

Highlights

Finite-sample theory for conformalized quantile regression under covariate shift
Evolve's verdict on this 2026 arXiv paper credits finite-sample Lp bounds for CQR interval length and coverage under known covariate shift, with matching minimax fixed-score benchmarks. Indexed keywords include the Le Cam two-point method, conditional coverage, and minimax lower bounds; the review calls the theory rigorous, honest, and assumption-bounded.
First controlled optimizer benchmark for discrete diffusion
The 2026 benchmark compares AdamW, Muon, and SOAP for diffusion-model training. Evolve rates it well-run and reproducible (high readiness) but flags that text-task budgets were only 3% of published training lengths — a stated scope limit on the strength of its conclusions.
Above-average reproducibility across the assessed catalog
Both reviewed papers carry explicit reproducibility ratings (medium and high), averaging 78/100 over the 2 assessed papers — the highest scored dimension in this snapshot and a signal of benchmark discipline and methodological transparency.

Timeline

2026arXiv posting of 'Conformalized Quantile Regression and Minimax Limits of Fixed-Score Calibration under Known Covariate Shift'; Evolve credits finite-sample Lp bounds paired with matching minimax fixed-score benchmarks.
2026arXiv posting of 'Optimizers for Diffusion Models: A Controlled Benchmark', described by Evolve as the first controlled optimizer benchmark for discrete diffusion (AdamW, Muon, SOAP).
2026Both papers enter the Evolve catalog with reproducibility readiness rated medium and high respectively, yielding a reproducibility mean of 78 over the 2 assessed papers.

Papers in the Evolve catalog

Frequent co-authors

Rustam Isaev×1 Anton Conrad×1 Denis Belomestny×1 Sergey Samsonov×1 Arman Bolatov×1 Egor Shulgin×1 David Li×1 Abduragim Shtanchaev×1 Sebastian U. Stich×1 Maxim Panov×1 Peter Richtárik×1 Martin Takáč×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.

Profile is name-resolved from public bibliographic data; no ORCID or biographical details were provided, so identity linkage beyond this record is not established.
The evidence base is two 2026 arXiv papers reviewed by Evolve, both with the researcher in a middle, non-corresponding authorship position, so individual contribution share cannot be inferred.
Reproducibility (78/100) is a mean over only 2 assessed papers — a small sample that may shift as more work is reviewed.
Output, impact, and activity dimensions are unscored because citation counts, publication counts, and per-year cadence were missing from the snapshot.