Machine Learning in Materials Science

Machine learning is a branch of artificial intelligence and can be used as an efficient approach to predict many different properties of materials and surface stuctures. The idea is to develop a machine learning model that learns effectively from information already available for thousands of other materials by creating statistically optimized relationships between the given data. Once the model is trained sufficiently, it can make predictions for new materials with almost the same accuracy as conventional computational methods, but in only a fraction of the time and with a fraction of the computational effort. Machine learning is therefore a useful tool to speed up the quest for improved and novel materials.


Bayesian Optimisation Structure Search (BOSS)


We coupled total energy atomistic simulation methods with the Bayesian Optimization artificial intelligence technique to learn complex energy landscapes, stable structures and favourable properties. This has allowed us to understand large-scale interface configurations between organic molecules and crystalline substrates used in thin-film electronic devices.

People involved: Milica TodorovićHenri Paulamäki


Machine Learning Methods for Spectra of Novel Materials


The aim of this project is to develop machine learning models based on kernel ridge regression and deep neural networks that utilize the abundance of already available theoretical and experimental spectroscopic data. The models will be able to make instant predictions of spectra at negligible computational cost, thereby greatly accelerating the spectroscopic analysis of chemical structures and the discovery of entirely new materials.

People involved:  Annika StukeMilica Todorović


The Novel Materials Discovery Laboratory


Our group is also involved in building the Novel Materials Discovery (NOMAD) Laboratory, which provides homogenized data for machine learning applications and and a notebook environment and software tools for running machine learning tasks on this data.

People involved: Lauri Himanen

Page content by: communications-phys [at] aalto [dot] fi (Department of Physics) | Last updated: 07.12.2017.