- This event has passed.
TYC Highlight Seminar: Atomic-scale machine learning: what do models compute?
23 March @ 12:00 pm – 2:00 pm
Venue: B10, Molecular Sciences Research Hub, Imperial College London, White City Campus
12:00 – 12:30 – Nan Wu, PhD student, Sophia Yaliraki Group
Atomistic graph learning in allosteric processes
12:30 – 13:30 – Michele Ceriotti, EPFL
Atomic-scale machine learning: what do models compute?
Over the past decade, machine learning (ML) techniques have become an indispensable tool in the atomic-scale modeling of materials – in the form of data-driven potentials, and more generally of surrogate models for all quantities that can be obtained by an electronic-structure calculation.
Applying machine-learning techniques to simulations has some interesting conceptual implications: if a ML model is to be able to predict the outcome of a physics-based calculation, it should have sufficient flexibility, and the appropriate mathematical structure, to reproduce the desired physical interactions and processes.
In this talk I am going to summarize an ongoing effort to better understand the structure of a broad class of ML frameworks that are routinely used in atomistic simulations, revealing their strengths and limitations. I will discuss how to extract physical insights from a critical analysis of the model performance, and how to improve the performance of models by incorporating physical-chemical priors.
I will punctuate this discussion with examples of recent applications of atomistic ML to different classes of materials, such as high-entropy alloys and ferroelectrics.