Periodic Set Transformer: Material Property Prediction from Continuous Isometry Invariants

Jonathan Balasingham, Viktor Zamaraev, Vitaliy KurlinOpenReview 2024 · Submission to ICLR 2024 · Rejected · OpenReview

Abstract

Material or crystal property prediction using machine learning has grown popular in recent years as it provides an accurate and computationally efficient replacement to classical simulation methods. A crucial first step for any of these algorithms is the representation used for a periodic crystal. While similar objects like molecules and proteins have a fixed number of atoms and their representation can be built based upon a finite point cloud interpretation, periodic crystals are unbounded in size, making their representation more challenging. In the present work, we adapt the Pointwise Distance Distribution (PDD), a continuous isometry invariant for periodic point sets, as a representation for our learning algorithm. While the PDD is effective in distinguishing periodic point sets up to isometry, there is no consideration for the composition of the underlying material. We develop a transformer model with a modified self-attention mechanism that can utilize the PDD and incorporate compositional information via a spatial encoding method. This model is tested thoroughly with and without the use of compositional information on a variety of crystal datasets including the commonly used crystals of the Materials Project.
BibTeX
@misc{balasingham2024periodic,
  author = {Jonathan Balasingham and Viktor Zamaraev and Vitaliy Kurlin},
  title = {{Periodic Set Transformer: Material Property Prediction from Continuous Isometry Invariants}},
  year = {2024},
  url = {https://openreview.net/forum?id=MNwXif6AWA},
  howpublished = {OpenReview},
  note = {Submitted to ICLR 2024},
}

Loading related content...

Figure