ML-IP 2021 is a Psi-k tutorial workshop aimed at young and early-career researchers who are interested in using machine-learning interatomic potentials (ML-IP) in their work and would like to gain a solid theoretical and practical grounding in the field before embarking on a research project.

ML-IP 2021 has taken place fully online, on 15-19 November 2021.

Both the organizers and the invited speakers are also early-career researchers well equipped to share their working experience on ML-IP in practical atomistic simulations with those interested in getting started in the field. The workshop format will consist of short introductory lectures to the topics, but the main focus will be the hands-on tutorials guiding participants through the application of these concepts to relevant toy cases and their own research.

The two main scopes of this tutorial workshop are:

  • to give young and early-career researchers a solid introduction in the basic scientific techniques of designing, fitting, and validating an ML-IP for a new system

  • to provide a platform for those interested in using machine learning potentials in their work to connect to those involved in developing methods for ML-IP, in order to accelerate the adoption of ML techniques in the wider atomistic simulation community.

We hope this platform will also provide an opportunity to enter the field for those who might be interested in pursuing method development and addressing the remaining unsolved problems in machine-learning potential development.

Topics & tools

The ML-IP 2021 workshop will focus on the most successful and robust ML

  • linear methods with high-dimensional descriptors (e.g. SNAP, MTP, aPIP)

  • nonlinear methods, which include both kernel-based (e.g. GAP) and neural network (NN) approaches.

Speakers will provide interactive tutorials that the workshop participants can run on a cloud-based service, or download and run on their own computers.

We will additionally provide time on a compute cluster for participants to experiment with the tutorials and try out their own research ideas over the course of the workshop.

We shall rely on widely used video-conference platforms, like Zoom, Gather, and Slack, as well as tools like Deepnote to help participants run the tutorials as easily as possible.

Online format

ML-IP 2021 will take place entirely online, which minimizes the risk that the workshop would need to be cancelled or modified in the case of heightened COVID-19 risk. It also opens the workshop to participation from a wider audience by eliminating travel and accommodation costs.

We will make a conscious effort to select participants from diverse backgrounds and experiences, in order to make the field more accessible. To minimize issues caused by participants joining from widely different time zones, we have scheduled 3-4 talks per day and will record the talks, with speaker and participant consent, has been made available and may be found in the Resources section.


Max Veit (he/him/his) is a scientist in the Laboratory of Computational Science and Modeling (COSMO), led by Prof. Michele Ceriotti, at EPFL, Switzerland, where his focus is on extending the efficiency and capability of machine learning potentials using advanced representations. He pursued his master's and PhD at the University of Cambridge, UK, under Prof. Gábor Csányi, where he focused on modelling both local and intermolecular interactions in molecular liquids using machine learning techniques.

Elena Gelžinytė is a PhD student working with Prof. Gábor Csányi at the Department of Engineering, University of Cambridge. Her research focuses on creating reactive force fields for small molecules and radicals with relevance to oxidative metabolism. Previously, Elena has studied Natural Sciences at the University of Cambridge and has explored predictive toxicity models in Prof. Jonathan Goodman's group as part of her Master's thesis.

Venkat Kapil is a research fellow at the Department of Chemistry, University of Cambridge. He is an "Early Postdoc Mobility Fellow" (SNSF) working with Prof. Angelos Michaelides on simulating the quantum nature of nuclei in atomistic simulations of molecular materials. He earned his PhD degree in Prof. Michele Ceriotti's group at EPFL, Switzerland.

Felix-Cosmin Mocanu is a postdoctoral researcher at the Laboratoire de Physique Théorique at École Normale Supérieure, in Paris, working with Francesco Zamponi and Ludovic Berthier on the properties of glassy systems at low temperatures. He did his PhD studies at the University of Cambridge with Prof. Stephen Elliott and Gábor Csányi on modelling phase-change memory materials with machine learning interatomic potentials.

Federico Grasselli is a Marie Skłodowska-Curie fellow in the Laboratory of Computational Science and Modeling (COSMO), led by Prof. Michele Ceriotti, at EPFL, Switzerland. He earned his PhD degree in Physics and Nanosciences from the University of Modena and Reggio Emilia, Italy, working on indirect excitons under the supervision of Prof. Guido Goldoni, and he has been postdoc fellow in Prof. Stefano Baroni's group at SISSA, Italy, where he contributed to advancements in the ab-initio theory of heat and charge transport and its application to planetary materials.