Hi and welcome! My name is Martin Uhrin, I have a mixed physics and computer science background and love to work at the intersection of the two. Of the things I’ve worked on the common theme is development of computational and mathematical methods that help us to understand, and ultimately predict, atomistic structures (be it via empirical potentials, structure prediction, rationalisation, data mining or other means).
The methods I develop are fairly application agnostic, but one of the things that motivates me is contributing to solutions that drive our green transition, particularly energy materials where I think materials science can make a big contribution. Recently, I was fortunate enough to do a postdoc with Jin Chang and Tejs Vegge at DTU doing research on metal-air batteries.
The current focus of my research is on the development of generative models that, in some sense, understand the structure/property relationship so well that they can propose brand new atomic structures likely to have some set of desired properties. An important component to this is the use of equivariant neural networks that allow us to perform deep-learning while obeying the symmetries of the relevant physical laws. I’m super happy to be working with the talented e3nn team to bring this powerful technology into my research.