报告摘要：

　　In this presentation, we will begin by providing an overview of the deep potential model, including its background, construction, concurrent learning data generation method, and successful applications. We will then highlight two recent advancements in the deep potential model: the pretraining model DPA-1 and the long-range interaction model DPLR. DPA-1 utilizes the attention mechanism to address fast training and generation issues in large-scale datasets. We will demonstrate the interpretability of the DPA-1 model by showcasing the visualization of the element type embedding. On the other hand, the DPLR model accounts for long-range interactions that can significantly impact the system's behavior, in addition to the short-range interactions modeled by the standard DP model. We will provide examples to showcase the effectiveness of both DPA-1 and DPLR models.

报告人简介：

　　王涵，北京应用物理与计算数学研究所研究员，博士生导师。2011年毕业于北京大学数学科学学院，获得理学博士学位；2011至2014年于柏林自由大学数学与计算机学院从事博士后研究。2014年加入北京应用物理与计算数学研究所，现任研究员。主要研究兴趣为分子动力学模拟中的多尺度建模与计算方法。与合作者发展了基于深度学习的原子间相互作用建模与计算方法，解决了传统方法精度和效率无法两全的困境，将第一原理精度分子动力学模拟规模推进至亿原子量级。