About this Abstract |
Meeting |
2025 TMS Annual Meeting & Exhibition
|
Symposium
|
Algorithms Development in Materials Science and Engineering
|
Presentation Title |
DiSCoVeR 2.0: incorporating structural similarity as a search criteria for new materials |
Author(s) |
Taylor D. Sparks, Andrew Falkowski, Sterling Baird |
On-Site Speaker (Planned) |
Taylor D. Sparks |
Abstract Scope |
Machine learning already enables the discovery of new materials by providing rapid predictions of properties to complement slower calculations and experiments. Recently, generative models have even begun suggesting entirely new crystal structures altogether. However, a reasonable criticism is that the generated structures are often very similar structurally and or compositionally to known materials. In many instances, we would prefer unusual structures or chemistries to open up new directions of research into underexplored families of compounds. Recently, we proposed the DiSCoVeR algorithm which used the Element Movers Distance as a novel compositional distance metric in order to encourage the discovery of unique chemistries. Here, we extend DiSCoVeR to also include a new metric for structural similarity based on GridRDF calculations so that users can explore the pareto front along predicted performance, structural uniqueness, and compositional uniqueness. |
Proceedings Inclusion? |
Planned: |
Keywords |
Machine Learning, Computational Materials Science & Engineering, Modeling and Simulation |