Victor García Satorras

Timeline

1. 2019

   Hybrid inference for Lorenz

   In 2019, he coauthors “Combining Generative and Discriminative Models for Hybrid Inference” with Zeynep Akata and Max Welling. The paper combines graphical inference with a learned inverse model structured as a graph neural network; its Kalman-filter experiments estimate a noisy chaotic Lorenz Attractor trajectory more accurately than learned or graphical inference alone.

2. 2021

   E(n)-equivariant graph networks

   At the 38th International Conference on Machine Learning in 2021, he is first author with Emiel Hoogeboom and Max Welling of the paper introducing E(n)-Equivariant Graph Neural Networks (EGNNs). The model is equivariant to rotations, translations, reflections and permutations; the authors test it on dynamical systems, graph autoencoders and molecular property prediction.

3. 2021

   Equivariant normalizing flows

   At NeurIPS in 2021, he and Emiel Hoogeboom, Fabian Fuchs, Ingmar Posner and Max Welling integrate EGNNs as a differential equation to create E(n) Equivariant Normalizing Flows. They describe it as the first flow that jointly generates molecule features and 3D positions.

4. 2022

   Equivariant diffusion for molecules

   At the 39th International Conference on Machine Learning in 2022, he is second author of “Equivariant Diffusion for Molecule Generation in 3D,” with Emiel Hoogeboom, Clément Vignac and Max Welling. Their E(3) Equivariant Diffusion Model (EDM) jointly denoises atom coordinates and atom types, and outperforms earlier 3D molecular generative methods.

5. 2023

   Diffusion and force fields

   In 2023, he coauthors “Two for One: Diffusion Models and Force Fields for Coarse-Grained Molecular Dynamics” with Marloes Arts and other researchers. Their diffusion model learns a coarse-grained force field from protein structures in molecular-dynamics simulations without force inputs, and improves results across several small and medium protein simulations.

6. May 2024

   PhD thesis in Amsterdam

   In May 2024, he defends his thesis “Inductive Biases for Graph Neural Networks” at the University of Amsterdam, with M. Welling as supervisor and H.C. van Hoof as co-supervisor. During his PhD, he works in the Amsterdam Machine Learning Lab (AMLAB) on message-passing algorithms and generative models for molecular modelling, hybrid inference and time-series forecasting.

7. 2025

   BioEmu-1 protein ensembles

   In February 2025, Microsoft makes BioEmu-1 open source. The deep-learning model generates thousands of protein structures per hour on one graphics processing unit, and he is among its authors; its paper appears in Science in July 2025.

8. As of October 2026

   Senior Researcher in Amsterdam

   As of October 2026, he is a Senior Researcher at Microsoft Research AI for Science. He works in its Amsterdam team on deep learning for molecular modelling and methods for computational catalysis design.