Young Researcher’s Workshop on Machine Learning for Materials Science

May 6, 2019 to May 10, 2019
Location : CECAM-FI


  • Milica Todorovic (Department of Applied Physics, Aalto University, Finland)
  • Adam Foster (Aalto University, Finland)
  • Patrick Rinke (Aalto University, Helsinki, Finland)





The influx of machine learning (ML) algorithms from computer science into computational materials science (MS) have led to developments of novel computational methodologies and opened up novel routes to addressing outstanding problems. Over the past decade, seminal works have shown how to employ ML algorithms in global structure search [1,2], to predict stabilities of molecules [3] and solids [4], or target materials with particular functional properties [5,6]. Much has become known about the performance and versatility of ML techniques such as kernel ridge [3,4] and Gaussian process [7] regression, neural networks [5,8], genetic algorithms [1] and Bayesian approaches [6], with recent advances too numerous to mention. Materials descriptors that facilitate ML for both molecules and solids are growing ever more sophisticated [9,10] and have dramatically improved the accuracy of ML predictions.

The successes of this rapidly developing research field have also opened many questions on how to further enhance current methodology. Is learning materials descriptors the best long-term strategy, and should ML techniques be combined in workflows for maximum benefit? While it is now possible to predict materials properties given the structure, how can we tackle the inverse problem of predicting chemically-meaningful material structures given desired properties?

The 2019 “Machine Learning for Materials Science” (ML4MS) workshop will bring together experts from computer science (CS) and researchers applying ML in computational physics or chemistry for focused discussions with the community. The key players employing ML methods in materials science will provide an overview of the cutting-edge applications; experts from the CS community, who are familiar with materials science problems, will introduce additional algorithms and techniques that could help overcome the shortcomings of current methodology and develop approaches beyond the state-of-the-art. Timetabled discussion sessions will be organised around key issues such as:
- generalising materials descriptors
- inverse learning problems and generative models
- latent space and reinforcement learning approaches
- quantifying uncertainty in ML predictions
- combined ML workflows




[1] A. R. Oganov and C. W. Glass, J. Chem. Phys. 124, 244704 (2006).
[2] L. J. Nelson, G. L. W. Hart, F. Zhou, and V. Ozolins, Phys. Rev. B 87, 035125 (2013).
[3] M. Rupp, A. Tkatchenko, K.-R. Müller, and O. A. von Lilienfeld, Phys. Rev. Lett. 108, 058301 (2012). 
[4] F. A. Faber, A. Lindmaa, O. A. von Lilienfeld, and R. Armiento, Phys. Rev. Lett. 117, 135502 (2016).
[5] E. O. Pyzer-Knapp, K. Li, and A. Aspuru-Guzik, Adv. Fun. Mat. 25, 6495 (2015).
[6] A. Seko, A. Togo, H. Hayashi, K. Tsuda, L. Chaput, and I. Tanaka, Phys. Rev. Lett. 115, 205901 (2015).
[7] A. P. Bartok, M. C. Payne, R. Kondor, and G. Csanyi, Phys. Rev. Lett. 104, 136403 (2010).
[8] J. Behler and M. Parrinello, Phys. Rev. Lett. 98, 146401 (2007). 
[9] H. Huo and M. Rupp, arXiv preprint arXiv:1704.06439 (2017).