Generative Materials
Conditional diffusion and inpainting models for crystal structure prediction and materials discovery.
AI for Science | Materials Discovery
Xinzhe Dai
PhD Researcher in Materials Science and Engineering at Cornell University
I develop machine-learning methods to understand and design complex materials—from generative crystal models to data-centric tools for atomistic simulation and electron microscopy.
Research focus
Scientific Machine Learning
Physically grounded learning for atomistic simulations, interatomic potentials, and crystalline structure analysis.
AI for Microscopy
Data-driven microscopy images and spectra denoising, atom localization, and local structural order deciphering.
Selected work
View all →
2025 · Published
Host-Guided Crystal Structure Generation
Conditional generative models for placing atoms inside symmetrized host structures, combining diffusion-based inpainting with foundation potentials for materials discovery.
2024 · Published
Generative Inpainting for Intercalation Materials
An SE(3)-equivariant score model that predicts intercalant positions in host crystals for energy-storage materials design.
2023 · Open Source
DeepSTEM: AI Copilot for Electron Microscopy
Open-source tools for STEM image and EELS analysis, including denoising, atom localization, chemical characterization, and structure discovery.
Recent publications
Full list →Mechano-diffusion of particles in hydrogels
Extreme Mechanics Letters
Crystal structure prediction with host-guided inpainting generation and foundation potentials
Materials Horizons
Practical approaches for crystal structure predictions with inpainting generation and universal interatomic potentials
AI4X 2025 International Conference