Publications

Crystal structure prediction with host-guided inpainting generation and foundation potentials

Published in Materials Horizons, 2025

This paper introduces a generative inpainting framework (CHGGen) that enables intercalation-based materials design by conditionally generating atomic positions within host crystal structures.

Recommended citation: Zhong, P., Dai, X., Deng, B., Ceder, G., Persson, K., Crystal structure prediction with host-guided inpainting generation and foundation potentials. Materials Horizons (2025) https://pubs-rsc-org.proxy.library.cornell.edu/en/content/articlelanding/2025/mh/d5mh00774g/unauth

Inpainting crystal structure generations with score-based denoising

Published in ICML 2024 Workshop AI4Science, 2024

This paper provides a way for intercalation-based materials design using generative inpainting method.

Recommended citation: Dai, X., Zhong, P., Deng, B., Chen, Y., Ceder, G. Inpainting crystal structure generations with score-based denoising. ICML Workshop AI4Science (2024) https://openreview.net/pdf?id=T1mIt5exUF

DeepSTEM: Copilot for STEM Z-contrast Imaging and STEM-EELS Data Processing and Analysis

Published in , 2023

An open-source AI copilot for STEM Z-Contrast imaging and STEM-EELS data processing and analysis.

Download here

Universal Interatomic Potential: Benchmarked with NequIP

Published in , 2023

How to train universal interatomic potential from scratch?

Low-dose imaging denoise with one pair of noise images

Published in Optics Express, 2023

This paper is about an unsupervised denoising method used for low-dose imaging.

Recommended citation: Yang, D., Lv, W., Zhang, J., Chen, H., Sun, X., Lv, S., Dai, X., Luo, R., Zhou, W., Shi, Y., Qiu, J. Low-dose imaging denoise with one pair of noise images. Optics Express (2023) https://www.researchgate.net/publication/370049095_Low-dose_imaging_denoising_with_one_pair_of_noisy_images

Data-centric Crystal Structure Identification in Atomistic Simulations

Published in , 2022

Post-processing of atomistic simulation data with GCNN and feature engineering.