NVIDIA Research · AI for Science

Valentin Duruisseaux

Accelerating scientific computing from simulation to design

I leverage computational mathematics and machine learning to develop scalable methods to
accelerate scientific simulations, computational discovery, and engineering design.

Research Scientist at NVIDIA Research in the AI-Aided Engineering group

Publications Google Scholar Semantic Scholar ResearchGate DBLP GitHub LinkedIn

AI-Aided EngineeringAIE develops foundational models and algorithms for physical systems. These models are trained on observational or simulation data to approximate complex physical processes efficiently, while preserving the geometry, physical conditions, and uncertainty needed for scientific and engineering use.

Explore AIE

Research

Accelerating science and engineering with physics-informed machine learning

My research brings together operator learning, numerical analysis, and scientific computing to accelerate high-fidelity scientific simulations. I develop surrogates that incorporate the relevant governing equations, geometry, symmetries, and constraints, while maintaining generalization across discretizations, resolutions, parameter settings, and operating regimes.

The resulting models complement traditional solvers with efficient, differentiable tools for simulation and inverse design in scientific and engineering applications.

Operator Learning

Learning operators directly on function spaces rather than through finite-dimensional vector discretizations
Discretization-consistent architectures across grids, resolutions, and geometries
Mathematical foundations, architecture development, and
open-source implementations

FNOs Explained From Networks to Operators NeuralOperator Library ↗

Physics-Informed Machine Learning

Incorporating governing equations, geometry, symmetries, and constraints in machine learning models.

Soft constraints through physics-based losses, with methods
for irregular geometries and nonperiodic domains
Hard constraints through projection and architecture modifications for exact enforcement

Fourier Continuation PINOs Hard Constraints Projection Mollified GNOsEuler Singularity Discovery Projected NDEs Equation RecastNearly-Periodic Symplectic Maps Lie Group Dynamics and Control

Accelerated Simulations

Fast, accurate surrogates for repeated, computationally intensive scientific simulations
Generalization across resolutions, parameters, geometries, and operating regimes
Applications across fluid dynamics, weather and climate, fusion, photonics, quantum systems, neuroscience, and medical imaging

Quantum Control Sequences Neuron Dynamics Nonlinear Photonics 3D MRI Reconstruction

Accelerated Design

Differentiable surrogate models for inverse design and control
Gradient-based optimization of physical parameters, geometries, and control inputs
Inverse design of photonic structures, quantum control sequences, and CFD geometries

Quantum Control Sequences Shape Optimization Photonic Waveguides

Professional experience

From computational mathematics research to AI-aided engineering

Present

NVIDIA Research · AI-Aided Engineering

Research Scientist

Neural operators for high-fidelity physical systems
Geometry and physical conditioning across CAD, meshes, boundaries, and operating conditions
Reliable generalization and calibrated uncertainty across discretizations and physical regimes
Differentiable surrogates for simulation, optimization, and inverse design

2024-2026

Caltech Computing and Mathematical Sciences · Anima AI+Science Lab

Postdoctoral Researcher

Mathematical foundations and practical formulations of neural operators
Physics-informed operators using Fourier continuation, exact differentiation, and hard constraints
Generalization across discretizations, resolutions, physical parameters, and unseen geometries
Fast surrogate models across numerous applications, including fluid dynamics, weather and climate, nuclear fusion, photonics, quantum physics, neuroscience, and medical imaging
Differentiable surrogate models for inverse design and control across photonics, quantum systems, neuroscience, and fluid dynamics
Numerical discovery and stability analysis of singularity formation in nonlinear PDEs
Open-source operator learning through the NeuralOperator library

Earlier research internships

Los Alamos National Laboratory, 2022 · Machine learning for fast, structure-preserving simulation of
nearly-periodic Hamiltonian systems Siemens, 2023 · Operator learning framework for reconstructing high-resolution scientific simulations
from coarse or sparse data

Education

Mathematical and physical foundations for interdisciplinary scientific computing

2018-2023

University of California San Diego

Ph.D. in Mathematics · Computational Science

Research under the supervision of Professor Melvin Leok
Geometric numerical integration and adaptive symplectic methods for accelerated optimization
Variational formulations and structure-preserving algorithms in vector spaces and on Riemannian manifolds
Structure-preserving dynamics learning for nearly-periodic Hamiltonian dynamics and controlled robots on Lie groups

Thesis ↗

2017-2018

University of Cambridge

Master of Advanced Study in Mathematics · Part III

Merit · Dissertation under the supervision of Professor Arieh Iserles on highly oscillatory quadrature

2013-2017

McGill University

Joint Honours B.Sc. in Mathematics and Physics

First Class Honours with Distinction · Minor concentration in Computer Science
Research on symplectic numerical integration of Hamiltonian systems with Professor Gantumur Tsogtgerel
Research on numerical bifurcation analysis of delay differential equations with Professor Tony Humphries