M. UHRIN Martin

Research activities

I work at the intersection of physics and computer science, developing computational and mathematical methods that help us understand—and ultimately predict—atomistic structure and behavior. Across my career, the unifying theme has been building tools that extract physical insight from complex materials data, whether through empirical potentials, structure prediction, ab initio modelling, or machine-learning–driven rationalisation.

My research is currently structured around three main axes:

1. High-throughput ab initio workflows for materials screening.
I design scalable computational pipelines that automate density-functional–theory calculations and enable rapid exploration of large materials spaces, supporting data generation, screening, and hypothesis testing at scale.

2. Advanced physics-informed and equivariant machine learning.
A major part of my work focuses on building symmetry-preserving neural networks that respect fundamental physical laws. These models allow us to extract accurate structure–property relationships with far less data, and I am delighted to collaborate closely with the e3nn team to push this technology forward.

3. Generative models for inverse design and characterisation.
My long-term goal is to develop generative models that “understand” materials well enough to propose entirely new atomic structures with targeted properties, as well as assist in interpreting experimental signatures. These models open the door to inverse design workflows that couple predictive ML with automated sampling of the vast chemical and structural design space.

Although the methods I develop are broadly applicable, I am particularly motivated by challenges central to the green transition. Energy materials—especially electrolytes for batteries—represent an area where advances in computation and machine learning can have real impact. Before joining my current lab, I had the great opportunity to work with Jin Chang and Tejs Vegge at DTU on metal–air batteries, which strengthened this focus.