Massive Atomistic Diversity (MAD) dataset
收藏资源简介:
MAD数据集是一种新型数据集,其创建目的是推动'通用性'的极限,适用于有机和无机原子级模拟的多种系统维度。数据集覆盖了广泛的配置空间,包含随机和非平衡结构,以适应复杂的原子模拟协议。参考电子结构计算具有高度一致性和鲁棒性,确保不同结构中的结构和化学基元以相同方式处理。该数据集旨在减少训练时间成本,同时保持代表性。MAD数据集被用于训练PET-MAD模型,该模型在多种材料类别的先进原子模拟中表现出色。
The MAD dataset is a novel dataset developed to push the limits of 'generality', applicable to multiple system dimensions for both organic and inorganic atomic-scale simulations. It covers a wide configuration space, including both random and non-equilibrium structures, to accommodate complex atomic simulation protocols. The reference electronic structure calculations exhibit high consistency and robustness, ensuring that structural and chemical motifs across different structures are treated uniformly. This dataset aims to reduce training time and computational costs while maintaining representativeness. The MAD dataset has been used to train the PET-MAD model, which delivers outstanding performance in advanced atomic simulations across various material categories.




