Basic theories and methods


We mainly focus on cross-scale simulation based on machine learning, and further promote the application of “AI for Science” in various micro-scale simulation.
Taking molecular dynamics simulation as an example, we have developed a series of methods and tools such as Deep Potential Molecular Dynamics (DeePMD). Through machine learning to fit the data generated based on first principles, DeePMD realizes the efficiency improvement of molecular dynamics with first-principles accuracy.


Based on “AI+Physics+HPC”, we pushed the limit of molecular dynamics with ab-initio accuracy to 100 million atoms in 2020, and won the Gordon Bell Prize.


DeePMD has become a popular research method in theoretical chemistry, computational physics, molecular biology, materials science, and more than one hundred papers have been published based on DeePMD.
Furthermore, in order to make the training of potential function more convenient and greatly save the computing resources in data production, we proposed DPA-1,a pre-training model,which greatly improved the capacity and migration ability of the model by adopting the attention mechanism.

Based on DPA-1, we further trained on ultra-large scale datasets, and obtained pre-training model covering nearly 70 chemical elements. By visualizing the elements information of the model, we found that it distributed spirally, and corresponded to the positions in the periodic table, which also proved the interpretability of the pre-training model.


Researchers have proved the superiority of the pre-training model in various material, such as alloys, semiconductors, solid electrolytes, etc., and a small amount of data can be used to obtain good model accuracy in the new scene, thus greatly reducing the amount of data and the cost required for training model.


We also launched the “AIS Square” project, which provides an open-source platform of data, software, models and workflows for “AI for Science” researchers in the world. It will serve as an important infrastructure to promote the establishment of “AI for Science” ecosystem, and accelerate the meta-innovation speed.

About Us



Our Team



Address:No.150 Chengfu Road, Haidian District, Beijing

Email :