Anima Anandkumar

California Institute of Technology

Unclaimed · AI-structured profileORCID OpenAlex

Scientific machine learning researcher at Caltech: neural operators for parametric PDEs and Bayesian inverse problems

papers here

reviewed by Evolve

first-author

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—

Not yet scored: the output axis would measure publication volume and authorship roles (first, middle, last positions) over the profiled window, but the underlying count data is absent from this snapshot, so no estimate is offered here.

Impact—

Not yet scored: the impact axis would capture citation counts, aggregate citation indicators, or downstream usage of the work. None of these signals are present in the provided data, so this profile cannot assess how widely the cataloged papers have been received.

Focus0/100

The focus dimension scores 0, reflecting the snapshot's field distribution: it splits evenly, 0.5 to 0.5, between 'scientific machine learning (neural operators for parametric PDEs)' and 'scientific machine learning / Bayesian inverse problems'. Both labels sit inside scientific machine learning and share neural-operator and inverse-problem machinery, so the split reads as two adjacent threads rather than unrelated agendas, but the even split is what the score records.

Collaboration61/100

The collaboration dimension scores 61. The snapshot records 20 distinct coauthors, and the Evolve catalog shows a recurring core among the twelve most frequent partners, including Guangyuan Wang, Mads Toftrup, Sebastian Loeschcke, Yixuan Wang, Qiyun Cheng, Valentin Duruisseaux, and Shaowu Pan. All three catalog papers are team efforts, with Anandkumar in the last author position on two and a middle position on one, none as corresponding author.

Activity—

Not yet scored: the activity axis would track publication cadence and year-by-year output over time. Only three 2026 catalog entries exist here, so the longitudinal data needed to judge how active this researcher is across years is insufficient.

Reproducibility100/100

The reproducibility dimension scores 100, computed as a reproducibility mean of 100 across 2 papers — the two Evolve-reviewed entries. Both 'PosteriorBench' and 'Equation Recast' carry a high reproducibility-readiness rating, consistent with their benchmark-centered designs. The unreviewed 'SS-ESOAP' is not in the average, so the perfect score rests on a small but favorably reviewed base.

Research summary

Anima Anandkumar (California Institute of Technology; ORCID 0000-0002-6974-6797) is profiled here from public bibliographic data and three arXiv papers in the Evolve catalog, all posted in 2026. The snapshot's field distribution splits evenly, 0.5 to 0.5, between scientific machine learning for parametric PDEs via neural operators and scientific machine learning for Bayesian inverse problems. 'Equation Recast for Canonical Operator Learning Across Parametric PDEs' (2026, middle author position) recasts parametric PDE learning as one canonical operator plus an analytic fixed-point correction; Evolve calls it elegant, well-benchmarked, and honest about limits, with statistics thin. 'PosteriorBench' (2026, last author position) moves evaluation of generative inverse solvers from point estimates to posterior matching, earning a solid verdict alongside pool-based-reference and author-overlap fairness caveats. A third entry, 'SS-ESOAP: Self-Scaled Adaptive Preconditioning for Physics-Informed Learning' (2026, last author position), has no Evolve review yet. The window records 20 distinct coauthors, including Guangyuan Wang, Yixuan Wang, and Shaowu Pan. Both reviewed papers rate high in reproducibility readiness, producing a reproducibility score of 100 from 2 papers; output, impact, and activity metrics are not yet available.

Collaboration pattern

The snapshot lists 20 distinct coauthors with a dense recurring core: Guangyuan Wang, Mads Toftrup, Sebastian Loeschcke, Yixuan Wang, Qiyun Cheng, Valentin Duruisseaux, Cesar F. Clauser, Md Hossain Sahadath, Huihua Yang, Shaowu Pan, Nathaniel Ferraro, and Wei Ji. All three 2026 catalog papers are collaborative — last-author position on PosteriorBench and SS-ESOAP, middle position on Equation Recast, and none as corresponding author — a pattern of senior technical contribution within team-led projects.

Research keywords

posterior matchinggenerative inverse solversdiffusion modelsBayesian inverse problemsneural operatorsuncertainty quantificationoperator learningparametric PDEsfixed-point iterationdefect correctionzero-shot extrapolationphysics-informed learning

Highlights

PosteriorBench: distributional benchmarking for generative inverse solvers
This 2026 arXiv paper moves evaluation of Bayesian inverse solvers from point estimates to posterior matching, spanning diffusion-based and neural-operator solvers with uncertainty quantification; Evolve calls it a solid bar for distributional benchmarking, while flagging pool-based reference posteriors and author-overlap fairness caveats.
Equation Recast: one canonical operator for parametric PDEs
This 2026 arXiv paper recasts learning across parametric PDEs as a single canonical operator learning problem plus an analytic fixed-point defect correction, supporting zero-shot extrapolation; Evolve judges it elegant, well-benchmarked, and honest about limits, though the statistical analysis is thin.
SS-ESOAP: self-scaled adaptive preconditioning for physics-informed learning
The third 2026 catalog entry proposes self-scaled adaptive preconditioning for physics-informed learning, with Anandkumar in the last author position; it has not yet received an Evolve review, so no verdict, keywords, or reproducibility rating are on record.

Timeline

2026Posts 'PosteriorBench: From Point Estimates to Posterior Matching in Evaluating Generative Inverse Solvers' to arXiv in the last author position; Evolve reviews it as a solid distributional benchmark with fairness caveats.
2026Posts 'Equation Recast for Canonical Operator Learning Across Parametric PDEs' to arXiv in a middle author position; the paper earns a high reproducibility-readiness rating.
2026Posts 'SS-ESOAP: Self-Scaled Adaptive Preconditioning for Physics-Informed Learning' to arXiv in the last author position; no Evolve review is on record yet.

Papers in the Evolve catalog

Frequent co-authors

Guangyuan Wang×1 Mads Toftrup×1 Sebastian Loeschcke×1 Yixuan Wang×1 Qiyun Cheng×1 Valentin Duruisseaux×1 Cesar F. Clauser×1 Md Hossain Sahadath×1 Huihua Yang×1 Shaowu Pan×1 Nathaniel Ferraro×1 Wei Ji×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 scores derive from OpenAlex-based public bibliometric data; the output, impact, and activity axes lack underlying data and are unscored.
The Evolve catalog contains only three arXiv papers from 2026; no peer-reviewed venue, citation, or career-history data is included.
The reproducibility score of 100 rests on just 2 reviewed papers; 'SS-ESOAP' is unreviewed and excluded from that average.
The profile is name-resolved from public bibliographic data (ORCID 0000-0002-6974-6797); no independent biographical verification is performed beyond that.