PCSL x AI Center Seminar - AI for Science series - Dr. Oliver Dicks

Event details
Date | 08.04.2025 |
Hour | 16:00 › 17:00 |
Speaker | Dr. Oliver Dicks |
Location | Online |
Category | Conferences - Seminars |
Event Language | English |
The talk is jointly organized by the PCSL Lab and the EPFL AI Center.
For on-site logistics, please use the following form to register (with your EPFL email address): Here.
The talk will be followed by a coffee session.
Host: Dr. Daniel Korchinski
Title
Finetuning Machine Learning Potentials for High Entropy Oxide Material Discovery
Abstract
MACE is an equivariant message passing neural network (MPNN) machine learning interatomic potential (MLIP) that achieves near quantum method calculation level accuracy for materials incorporating nearly the whole periodic table of elements for a fraction of the computational cost. This makes it ideal for exploring the huge compositional space of the novel class of materials known as high entropy oxides (HEOs). However, despite being trained on a database of over 1 million material systems whose properties have been calculated using density functional theory, there are still regions of chemical and structural space where its accuracy is insufficient to estimate chemical stability. We demonstrate methods to construct minimal training databases to finetune MACE for specific target systems and elements to aid in high-throughput structure search for brand new synthesizable materials not achievable with traditional methods, and then subsequently calculate their properties with a high degree of accuracy.
Bio
Oliver Dicks is a computational physicist specializing in condensed matter theory and material discovery. He holds a Marie Sklodowska-Curie (UKRI Guarantee) postdoctoral research fellowship at the University of British Columbia’s Quantum Matter Institute researching the use of machine learning potentials to model a new class of materials known as high entropy oxides. He has worked to develop methods to calculate superconducting properties in high-throughput structure search applications, in collaboration with Intellectual Ventures. His additional scientific interests include the effects of radiation damage on glass storage mediums for nuclear waste, as well as liquid matter theory and density functional theory. When not doing physics, he enjoys running, climbing, and skiing through the mountains of Canada.
Links
Practical information
- General public
- Free