Abstract

We will present some examples of recent work carried out in our group at Aalto University using machine learning techniques to study the structure and properties of materials. Patricia will present a new approach to low-dimensional embedding based on combining data classification (clustering) with traditional embedding techniques (multidimensional scaling), enhanced with sparsification to handle large databases. She will show an application of the method to carbon-based materials. Miguel will then talk, first, about adding van der Waals support to GAP force fields by learning the Hirshfeld effective volumes, a method that results in very little overhead over a regular GAP calculation and, second, about predicting X-ray photoelectron spectra from a multiscale model based on DFT and GW data, all glued up with machine learning based regression.