A Diffusion-Based Pre-training Framework for Crystal Property Prediction

Authors: Zixing Song, Ziqiao Meng, Irwin King

AAAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments demonstrate that Crys Diff can significantly improve the performance of the downstream crystal property prediction task on multiple target properties, outperforming all the SOTA pre-training models for crystals with good margins on the popular JARVIS-DFT dataset.
Researcher Affiliation Academia Zixing Song, Ziqiao Meng, Irwin King The Chinese University of Hong Kong zxsong@cse.cuhk.edu.hk, zqmeng@cse.cuhk.edu.hk, king@cse.cuhk.edu.hk
Pseudocode Yes Algorithm 1: Pre-training Phase of Crys Diff
Open Source Code No No explicit statement or link providing access to open-source code for the described methodology.
Open Datasets Yes We collect 800K untagged crystal graph data from two popular materials databases, Materials Project (MP) (Jain et al. 2013) and OQMD (Saal et al. 2013), to pre-train the Crys Diff model. Further, to evaluate the fine-tuning performance of Crys Diff compared with other crystal property predictors, we select the 2021.8.18 version of JARVIS-DFT (Choudhary et al. 2020), another popular materials database, for the downstream property prediction task.
Dataset Splits Yes For each property, all the models are trained on 80% data, validated on 10%, and tested on 10% of the data.
Hardware Specification No No specific hardware details (e.g., GPU/CPU models, memory) are provided for running experiments.
Software Dependencies No No specific software dependencies with version numbers (e.g., Python version, library versions) are listed.
Experiment Setup No The paper describes the data splits and loss functions, but does not provide specific hyperparameter values such as learning rate, batch size, number of epochs, or optimizer settings for training.