Jean Kossaifi — Group Lead · AI‑Aided Engineering · NVIDIA Research
My work spans computer vision, machine learning, tensor methods, and neural operators. At NVIDIA Research, I lead the AI-Aided Engineering group, developing methods and open-source systems for simulation, prediction, and design.
About
I develop AI methods and systems that help engineers and scientists accelerate simulation, prediction, and design. The aim is to preserve the structure of physical problems, their geometry, conditions, and regimes—while making computation fast enough to change how ideas are explored.
Today I lead the AI-Aided Engineering group at NVIDIA Research, where we advance Physics AI across engineering, weather, and materials. Open source is central to that work: I created and lead TensorLy and NeuralOperator, making advanced tensor and operator-learning methods easier to use, extend, and apply.
Before NVIDIA, I was a founding member of the Samsung AI Center in Cambridge. My academic training spans mathematics, computer science, and AI. I earned a French engineering degree (Diplôme d’Ingénieur) in mathematics, computer science and finance, alongside a BSc in Advanced Mathematics. I then earned an MSc in Advanced Computing with Distinction from Imperial College London, followed by a PhD in Artificial Intelligence at Imperial under the supervision of Professor Maja Pantic in the i.bug group.
Signature work
Open-source software
Democratizing tensor methods.
I created TensorLy to make tensor methods easier to use and extend. It provides a consistent Python API for tensor decomposition, regression, and tensorized neural networks, and is used across research and industry.
Industry impact
Probabilistic medium-range weather forecasting.
I led the development of ATLAS, a transformer that produces large forecast ensembles orders of magnitude faster than traditional numerical weather prediction.
Open-source software
Operator learning for PDE and scientific-computing problems.
I co-lead NeuralOperator, an open-source PyTorch library for training neural operators. Part of the PyTorch Ecosystem, it provides tested implementations and training utilities for operator learning.
Open-source software
Easy configuration of Python projects with minimal boilerplate. A clean, Pythonic approach to configuration management, particularly adapted for deep learning research.
Research agenda
Learning mappings between function spaces so models can transfer across resolutions, geometries, and physical regimes.
Building architectures that respect irregular domains, boundary conditions, and the structure of engineering problems.
Using multilinear structure to make large learning systems more compact, expressive, and computationally efficient.
Turning fast learned surrogates into useful tools for optimization, exploration, and real engineering decisions.
Research in practice
The goal is not a faster benchmark in isolation. It is a model that preserves physical structure, generalizes across design changes, and becomes useful inside the engineering loop.
G maps the input function a to the solution u
Selected publications
Sebastian Loeschcke, David Pitt, Robert Joseph George, Jiawei Zhao, Cheng Luo, Yuandong Tian, Jean Kossaifi, Anima Anandkumar
Transactions on Machine Learning Research
Jean Kossaifi, Nikola Kovachki, Morteza Mardani, Daniel Leibovici, Suman Ravuri, Ira Shokar, Edoardo Calvello, Mohammad Shoaib Abbas, Peter Harrington, Ashay Subramaniam, Noah Brenowitz, Boris Bonev, Wonmin Byeon, Karsten Kreis, Dale Durran, Arash Vahdat, Mike Pritchard, Jan Kautz
arXiv:2601.18111
Kamyar Azizzadenesheli, Nikola Kovachki, Zongyi Li, Miguel Liu-Schiaffini, Jean Kossaifi, Anima Anandkumar
Nature Reviews Physics
Jean Kossaifi, Nikola Kovachki, Zongyi Li, David Pitt, Miguel Liu-Schiaffini, Robert Joseph George, Boris Bonev, Kamyar Azizzadenesheli, Julius Berner, Valentin Duruisseaux, et al.
arXiv:2412.10354
Updates
Academic leadership & service
Current roles supporting the research community and its institutions.
2026 and 2025
Organizing Committee — Lead Communication Chair (2026), Communication Chair (2025)
2024–present
Editorial Board — Action Editor
2024
Editor — Special Topic 'Tensor Methods for Deep Learning'
Contact
For research conversations, open-source collaboration, and speaking enquiries, email is the best place to begin.
Jean Kossaifi — Group Lead · AI‑Aided Engineering · NVIDIA Research