DPA-2: a large atomic model as a multi-task learner

• Published in: npj computational materials 2024-12, Vol.10 (1), p.293-15
• Main Authors: Zhang, Duo, Liu, Xinzijian, Zhang, Xiangyu, Zhang, Chengqian, Cai, Chun, Bi, Hangrui, Du, Yiming, Qin, Xuejian, Peng, Anyang, Huang, Jiameng, Li, Bowen, Shan, Yifan, Zeng, Jinzhe, Zhang, Yuzhi, Liu, Siyuan, Li, Yifan, Chang, Junhan, Wang, Xinyan, Zhou, Shuo, Liu, Jianchuan, Luo, Xiaoshan, Wang, Zhenyu, Jiang, Wanrun, Wu, Jing, Yang, Yudi, Yang, Jiyuan, Yang, Manyi, Gong, Fu-Qiang, Zhang, Linshuang, Shi, Mengchao, Dai, Fu-Zhi, York, Darrin M., Liu, Shi, Zhu, Tong, Zhong, Zhicheng, Lv, Jian, Cheng, Jun, Jia, Weile, Chen, Mohan, Ke, Guolin, E, Weinan, Zhang, Linfeng, Wang, Han
• Format: Article
• Language: English
• Subjects: 639/301/1034/1035; 639/301/1034/1037; Artificial intelligence; Characterization and Evaluation of Materials; Chemistry and Materials Science; Computational Intelligence; Developmental stages; Electronic structure; Materials Science; Mathematical and Computational Engineering; Mathematical and Computational Physics; Mathematical Modeling and Industrial Mathematics; Modelling; Molecular modelling; Potential energy; Theoretical
• ISSN: 2057-3960
• DOI: 10.1038/s41524-024-01493-2

The rapid advancements in artificial intelligence (AI) are catalyzing transformative changes in atomic modeling, simulation, and design. AI-driven potential energy models have demonstrated the capability to conduct large-scale, long-duration simulations with the accuracy of ab initio electronic structure methods. However, the model generation process remains a bottleneck for large-scale applications. We propose a shift towards a model-centric ecosystem, wherein a large atomic model (LAM), pre-trained across multiple disciplines, can be efficiently fine-tuned and distilled for various downstream tasks, thereby establishing a new framework for molecular modeling. In this study, we introduce the DPA-2 architecture as a prototype for LAMs. Pre-trained on a diverse array of chemical and materials systems using a multi-task approach, DPA-2 demonstrates superior generalization capabilities across multiple downstream tasks compared to the traditional single-task pre-training and fine-tuning methodologies. Our approach sets the stage for the development and broad application of LAMs in molecular and materials simulation research.