Geometric Modeling of Crystal Structures Using Transformers

Speaker: Tatsunori Taniai
Affiliation: OMRON SINIC X Corporation
Homepage: https://taniai.space/
Date and Time: Thursday, June 5, 2025, 15:00–16:00 (JST)
Location: AIP Open Space and Zoom (hybrid format)
AIP Open Space: *only available to AIP researchers.

Abstract

Materials are fundamental to modern life—serving as semiconductors in electronic devices, magnets for everyday use, and active components in lithium-ion batteries. At the atomic scale, these materials exhibit a periodic arrangement of atoms in three-dimensional space, a key property that influences their physical and chemical behavior.

Historically, the relationship between crystal structures and material properties has been studied through a combination of experiments and physics-based simulations. Recently, machine learning has emerged as a powerful tool to further advance this understanding.

In this talk, I will present our recent research introduced at ICLR 2024 and 2025, focusing on transformer-based geometric encoders designed for crystalline materials. These models aim to learn from the periodic 3D geometry of crystal structures to predict various material properties. I will also highlight potential extensions of this approach to broader applications, including unsupervised representation learning and generative modeling of materials.